TW200951832A - Universal lookup of video-related data - Google Patents

Abstract

A universal video-related lookup system and method receives a request for information associated with specific video content from a requesting device. The system and method identify a first video content identifier associated with the specific video content and retrieves first metadata associated with the specific video content based on the first video content identifier. Next, the system and method translate the first video content identifier into a second video content identifier associated with the specific video content and retrieves second metadata based on the second video content identifier. The first metadata and the second metadata are then provided to the requesting device.

Description

200951832 VI. Description of the Invention: [Technical Fields of the Invention] A system for finding video-related data and a method thereof, in particular, a device and method for determining and correlating a complex number with a specific visual display, and a method thereof An apparatus for aggregating metadata from one or more metadata sources and methods thereof, the metadata being associated with a particular video content. [Prior Art] 0

The particular video content at the video store may have a point of view that is directed to the video and is only relevant to the video. This type of job is usually _ a Globally Unique identifier (GUID). Examples of the video storage provided by the present invention are distributed websites including video hosts (such as NetFlix, Y〇uTube, or Huh, etc.), collections of DVD media, collections of multimedia files, or peer-to-peer (peer-to- Peer) network. In particular, the GUID in each video store is unique. For example, the video in the online core _Y〇uTube has a _uniform resource locator (URL) ’. The network address is unique in the video resource on the line. The same 'slot case; BitTorre_ peer-to-peer network has a hash value' that is calculated from the contents of the file and used as the identifier of the file. A DVD disc can be uniquely identified based on a hash value generated by a multimedia file recorded on a disc, usually by a DVDid. 4 In addition, there is a storage location for multiple video related information (also called video related metadata) in the storage of video content. Some examples include Wikipedia, which contains a detailed description of the movie plot and characters, a network movie material 200951832 library containing a list of actors in the movie (such as International Movie Database, IMDB), and subtitle websites (such as OpenSubtitles) that contain subtitles in different languages. And a website database (such as DVDXML database) containing information about DVDs. Many of these metadata stores are everywhere and use different types of identifiers as indexes. For example, DVD related information (such as: cover, chapter list, title, etc.) can be retrieved using the DVDid, and the subtitles in the subtitle website database can be retrieved according to the movie hash value, and the identifier is captured. Use peer-to-peer online at mtT〇rrent. Other sources of silk also index video related information by bribery movies or other hash-related values. Some online newsletters use ^ as an index for video-related information. Although there are many ways to identify different types of video content, these identifiers are irreplaceable. The Metadata Store typically uses some types of identities to access the source of the identifiable content (iv) (eg, the subtitles and movie hash values, as well as DVD information and DVDids, etc.). However, the identifier of one source of information will not be able to effectively confirm the video content in other sources. SUMMARY OF THE INVENTION The present invention discloses an apparatus and method thereof, wherein: the method for issuing a bribe disclosure includes: at least: receiving, by the requesting device, a request corresponding to the temple video content; the self-viewing source confirmation and the specific video content Corresponding first-video content identifier; according to the first video content identification ^ corresponding frequency content corresponding to the meta-data; conversion first-view_容朗符 is the first: the content of the language; according to the second vision And the specific video metadata; providing the first metadata and the second metadata to the requesting method 200951832, the method disclosed in the present invention, the method comprising the steps of: confirming the first video sequence related to the video program; and confirming the first related to the video program a second video sequence; calculating a correspondence between the first video sequence and the second video sequence _; determining a first video sequence = - and a calibration line of the second video sequence; storing a correspondence between the first video phase and the second video sequence: ; storing information of the calibration line between the first video sequence and the second video sequence. The method of the present invention comprises the steps of: receiving a request for metadata associated with a video program, requesting a first video content identifier corresponding to the corresponding video program; confirming a video sequence in the video program; and confirming the correspondence based on the video sequence The video program identifier-video content identifier; the metadata data is retrieved from the metadata source based on the second video content identifier; and the corresponding relationship is confirmed by the time axis of the corresponding video program and the time axis of the corresponding metadata. The method of the present invention includes the steps of: inputting a first video identifier, the first video identifier comprising a content identifier and a first coordinate in the first video content; and confirming the corresponding second video content a set of second video identifiers, wherein φ the second video content is similar to the first video content; performing the following steps for each of the set of second video identifiers: confirming a set of coordinates in the second video content; • Outputting the set of coordinates in the second video content to the request element corresponding to the first video content. The device disclosed in the present invention comprises: a communication module, configured to receive a request for a specific video content from a requesting device; and a processor coupled to the communication module to confirm a first video content identifier corresponding to the specific video content And searching for the first meta-data corresponding to the specific video content according to the first video content identifier, and converting the first video content identifier to the second video content identifier 'and according to the second video content identification 5 200951832 The second element data is used to provide the first element information and the second element request device. The device disclosed in the present invention includes at least: a communication module, configured to receive a first video identifier, a first video identifier, a content identifier, and a first coordinate in the first video content, to transmit a second video The second coordinate of the group corresponding to the second and the second object, the second coordinate of the group is in the second video content; the processing benefit is connected with the communication module for capturing the second video from the table An identifier for confirming the second video identifier, the table containing the first block for storing the video identifier, and the video identifier associated with the video valley similar to the view size of the first side of the test The second stop. The apparatus and method disclosed in the present invention differs from the prior art in that the present invention confirms the first video content identifier corresponding to a specific video content by receiving a request for the corresponding specific video content transmitted by the requesting device, and Retrieving the first meta-data corresponding to the specific video content according to the first video content identifier, and then converting the first video content identification to the specific video content corresponding to the second video identifier, and according to the second video content identifier After extracting the second element data, the first element material and the second element data request device are provided, and the problems existing in the prior art are overcome by the problem, and the technical effect of reducing the load can be achieved. [Embodiment] Hereinafter, the features and embodiments of the present invention will be described in detail in conjunction with the drawings and the embodiments, which are sufficient to enable the skilled in the art to fully understand the technical solutions of the present invention. Implementation, whereby the achievable effects of the present invention are achieved. The apparatus and method of the present invention determine and correlate a plurality of video content identifiers for a particular video content pair 200951832. Moreover, the apparatus and method of the present invention aggregates metadata corresponding to a particular video content from one or more metadata sources. For example, the system and method determines multiple versions of a particular video program, regardless of transcoding format, 'aspect ratio, commercial, modified program length, and the like. The phase 1 of the video content is represented by space and or time coordinates. "Figure 1" shows an example of an environment 100 in which the apparatus and method of the present invention can be applied. Environment 100 includes a plurality of video sources 102, 104, and 106 coupled to network 108. Video sources 102, 1〇4, and 1〇6 can be any form of video source storage, peer-to-peer network, computer system, or other system that has the ability to store, manipulate, transmit, or otherwise provide video content. . The video source_ sub contains a download or φ stream (four) program, a peer-to-peer network that provides video content exchange, a PC system with video content, a Digital Versatile Disc (DVD) player, Blu-ray Disc (Blu_rayDisc, BD) Players, digital video recorders, game consoles, Y〇u Plus^, NetFlix, BitTorrent and more. Examples of video content include movies, television shows, home video, computer-generated movies, and a complete movie or television program. Network 108 is a data communication network that uses any communication protocol and any type of communication medium. In one embodiment, the network 1〇8 is the Internet, (Intemet). In other embodiments, network 108 is a combination of two or more networks. Network 108 may be accessed via a wired or wireless communication network. The % context 100 also contains a plurality of metadata sources 11(), 112, and 114 that are connected to the network 108. Metadata sources 110, 112, and 114 can be in any form of metadata repository, computer system, or other system that has the ability to store, retrieve, or otherwise provide metadata related to video content. Video related meta-data 7 200951832 Contains information: DVD cover, subtitles: New, visual _ rong title, video program actor, video content story outline, video content author / producer and view _ I am concerned about the views of the visitors and so on. Examples of source (4) sources include the dvdxml.com database for web services, the 0pensubtitie org database, the users of ιμ〇β, and YouTube users. The dvdxml c〇m database contains information about the DVD version of the movie, such as the title and key interpretations. 〇pensubtitle, the database also contains subtitle files in a variety of different languages for various movies. In addition, environment 100 includes a universal video lookup system 116 and video device 118 that are coupled to network 1-8, respectively. The universal video lookup system 116 as described above identifies and associates a plurality of video content identifiers associated with particular video content. In addition, the universal video lookup system 116 aggregates metadata associated with particular video content from one or more metadata sources 11〇, 112, and 114. Details related to the components and operation of the universal video lookup system 116 will be described herein. Video device 118 is capable of receiving and processing video content, for example, on one or more display devices (not shown). Specific examples of the video device 118 include a computer, a set-top box (seU〇pb〇x, STB), a satellite receiver, a DVD player, a Blu-ray player, a digital video recorder, a game console, and the like. Although "Figure 1" shows three video sources 1, 2, 1〇4, and 106 and three meta data sources 110, 112, and 114', in reality, the specific environment 1〇〇T contains the number of thoughts. Video source and any number of metadata sources. In addition, the particular environment 100 can also include any number of video devices 118, a universal video lookup system 116, and other devices or systems (not shown) that are interconnected by the network 108. Although the various components and systems shown in "Figure 1" are connected to the network 1〇8 200951832, the universal video lookup system 116 can be connected to one or more components or systems via other networks or communication lines. . Fig. 2 shows an example of the implementation of the universal video search system 116 of the present invention. The universal video search system 116 includes a communication module 202, a processor 204, a video analysis module 206, and a storage device 208. The communication module 202 is responsible for communicating data and other information between the universal video discovery system 116 and other devices such as video sources, metadata sources, video devices, and the like. The processor 2〇4 executes various instructions necessary for the operation of the universal video search system ❹ U6. For example, the processor 204 can execute some methods and programs to process the video content identifier and the metadata related to the metadata provided by the present invention. content. Video analysis module 206 executes various video programs and video analysis instructions. For example, video analysis module 206 can identify the content contained in the displayed video material. The storage device 208 stores the data and other information used by the general video lookup system 116 during operation. The storage device 2〇8 may contain one or more volatile and/or non-electrical memory. In a particular embodiment, the storage device 208 includes a hard disk (HDD) that combines electrical and non-electrical memory.

The universal video search system 116 also includes a video DNA module 21, a corresponding module 212, a hash calculation module 214, and an advertisement management module 21 for video DNA. The module 210 identifies objects in the video content, thereby finding Correspondence between different video sequences. The corresponding module 212 analyzes a plurality of video sequences to find a spatial and/or temporal correspondence between the two or more video sequences. The hash calculation module 214 calculates the #段四目__ function of the video content material (_ such (8). The hash function is a calculation that converts a large amount of data into a smaller "heavy value (_★)". The hash value is typically used as an index for the data sheet (dirty) or other large amount of resources. The advertisement management module 216 performs various advertisement related functions, for example, the advertisement management module 216 can be used in various Based on the factors, multiple advertisements are selected and inserted into the video content. Although not shown in "Picture 2", the components constituting the universal video search system 116 are transmitted through one or more communication lines, such as a bus, and Other components within the general video search system 116 communicate. In a particular embodiment, the modules represented by "Fig. 2" (video analysis module 2〇6, video DNA module 21〇, corresponding mode) The group 212 hash calculation module 214 and the advertisement management module 216) can represent that a computer readable command is executed, for example, the processor 2 〇 4 executes a computer readable command. "Fig. 3" is a captured video. Content identifier and A flowchart of an embodiment of a meta-data corresponding to a specific video content. First, the program 300 receives a request for a message corresponding to a specific video content (step 3〇2). The request is as shown in FIG. The illustrated video device 118 receives the requested message. The requested message may include metadata or other versions of the video content corresponding to the video content. Next, the private sequence 300 confirms the video content identifier corresponding to the particular video content (step 304). The request may include a video content identifier corresponding to the particular video content, a link to the video content, or a portion of the particular video content itself. If the request does not include a video content identifier, the program 300 will be described as follows. The corresponding video content identifier is confirmed according to the specific video content. Then the program 300 retrieves the meta-data corresponding to the specific video content according to the video content identifier (step 306). The extracted meta-data can correspond to all the video content. Corresponding to a specific time interval in the video content or corresponding to a spatiotemporal object in the video content. Metadata can be extracted from any number of metadata sources 200951832, for example, the first source provides subtitle information, the second source contains employee information, and the third source contains comments corresponding to the video content. The user scores. Then, the program 300 converts the video content identifier corresponding to the specific video content into other video identifiers corresponding to the previously determined video content identifier (step 308). The video content identifier corresponds to The conversion of the video identifier can be executed in advance and stored in a database, data table or other data structure for subsequent retrieval. Various functions can be processed by the silk video to convert the above video content identifier into the corresponding video identifier. program of. For example, the video content identifier can be converted to a binary identity identifier, or a second video content identifier that is converted to video content within a time interval. In another implementation, the visual job is converted to a time-space object in the video content. For example, storing a pre-computed video content job (four) material table, the data table includes a first block and a second place, wherein the video content identifier is stored in the first field, and the first place

The video content of the video content is similar to the video content of the similar (or smoke) video corresponding to the video content identifier is included in the second field. Next, the program 300_ is associated with each of the financial resources (step 310). In this way, with the Yuanxiao sister *., the stomach tears should be the same - the various versions of the video content are from any number of sources. Finally, the meta-lean = information corresponding to the video content is sent to the requesting device (step 312). This condition allows the device to be viewed in part or all of the devices. In addition, the requesting device can be viewed by the user of the use of the oyster device, and from the available version of the ^^_€_= 里内谷, then the selection is displayed. Fig. 4 is a flow chart showing a procedure 400 of an embodiment for judging the correspondence of 200951832 of a plurality of video sequences corresponding to the same video program. These video sequences are typically different versions of the same video program. Different versions of the same video program may have different aspect ratios, different transcoding formats, or different broadcast versions (eg, a full-length movie version without commercials and a commercial-advertised advertisement for television broadcasts) version). First, the routine 400 determines a first video sequence column corresponding to the video program (step 402). The first video sequence can represent all or part of the video program. Next, the program 400 confirms the second video sequence corresponding to the same video program (step 404). As mentioned above, the first video sequence and the second video sequence are different versions of the same video program. Thereafter, the routine 400 calculates a correspondence between the first video sequence and the second video sequence (step 406). This correspondence may contain time coordinates and/or space coordinates. Many systems and methods can be used to calculate the correspondence between the first video sequence and the second video sequence. When two video sequences require time synchronization, it is particularly important to calculate the temporal correspondence of the two video sequences. For example, assuming that the subtitles included in the first video sequence are displayed along with the video content of the second video sequence, the subtitle should be displayed at the correct time of the second video sequence. In this case, the calculation of the time correspondence will provide the appropriate synchronization. When confirming the information of the geometric pattern in the video sequence, for example, identifying many objects in a mirror and a head, it is particularly helpful to calculate the spatial correspondence between the two video sequences. Different versions of a video program can have different resolutions, aspect ratios, etc., so that there must be a spatial correspondence between the exchanges between the two video programs. For example, a spatial correspondence can provide different types of content as well as metadata exchanges associated with different versions of the video. The program 400 in Fig. 4 then determines the calibration line (time and/or space) of the first video sequence and the second video 12 200951832. A plurality of alternative embodiments calibrate the first video sequence and the second video sequence based on calibrating the audio based data segments. The 'thief' program lion stores the correspondence between the calculated first video sequence and the second video sequence (step 410). The information of the stored correspondence relationship can be provided for subsequent follow-up in the two __ 解 财 ,, without the need to calculate the corresponding relationship. Finally, the routine 400 stores information relating to the ❹ calibration line of the first video sequence and the second video sequence (step 412). The information of the stored calibration line can be used to calibrate the two video phases in the future, and the calibration of the video sequence is required. The program 4 in "Fig. 4" judges the opposite side of the two special and frequency sequences corresponding to the same video program. The program repeats the steps for each pair of video sequences to pre-calculate the information and establish a poor library (or data structure) containing information about many relationships. A database of correspondence information is used to accelerate the operation of the -6 n6, avoiding the calculation of the correspondence contained in the database. The repository of information for the correspondence may be included in the universal video lookup system 116 or accessed by the communication video lookup system 116 (or other system) via the network 108 or other data communication line. In a particular embodiment, the information of the pre-established correspondence is provided by a data service to a plurality of systems or devices, such as video device 118. Fig. 5 is a flow chart showing a procedure of an embodiment for determining, capturing, and calibrating caption information from a movie played on a DVD. In the example of "Fig. 5", a request is received in order to obtain specific caption information related to a specific movie (step 5〇2). For example, a user may wish to watch a movie on a DVD disc, but Xi 13 200951832 hopes that the subtitle information is a special language, such as Russian. The dv〇 disc may contain subtitle information in English and Spanish' but does not contain Russian. Using the universal video search system 116 and program provided by the present invention, the user can watch movies on the DVD in Russian subtitles. Then, the program confirms the first motion identifier corresponding to the DVD disc (step, step 5〇4). For example, the 'first excitation identifier' may be a hash value generated by a hash identification calculation or a partial video content on a surface disc. In a special embodiment, it can be confirmed according to the difficult name corresponding to the visual capacity, the file-based hash value, or the hash value based on the Lang capacity, and the visual axis capacity (4) is stored in the DVD disc. The video content on it) the associated identifier. An example of a file name identifier is to use a URL of a fixed video store associated with video content that is permanently stored and not changed. The hash value of the caseline can be used to confirm the age of the same file in the point-to-point route, even if the independent case is recorded between the users, the file-based hash value such as "M〇viehash". Content-based hash values to identify your video content, such as the video DNA described previously. Therefore, the value of the Langrong silk is the constant file name, encoding program, processing of video content, or editing of video content. After confirming the DVD-ID identifier, the program 500 confirms the video sequence (step 506) in a movie stored on the DVD disc, such as the first 3 seconds of the movie or any other video sequence in the movie. Then, the program 5 confirms the second DVD identifier associated with the DVD based on the confirmed video sequence (step 5〇8). The second dvd identification code is selected based on the additional information required (such as video metadata). In this example, the second dvd identifier is selected based on the index type of the message used to confirm the source of the subtitle. Thus, the second DVD identifier is used to find and retrieve information on Russian subtitles in the subtitle source associated with the particular 200951832 DVD disc selected by the user (step 510). The post-employment process 500 determines the correspondence between the time axis of the DVD movie and the day/inter-axis of the power associated with the source of the subtitle (step 512). The subtitle information is then mapped to the DVD movie in accordance with the scale axis of the calibration' (step 514). This correspondence and calibration line is a display of subtitles that require time to synchronize the appropriate movie content. In a particular embodiment of the procedure 5, the correspondence between the time axis of the DVD movie and the time axis of the electrical shadow associated with the source of the subtitle is pre-computed and stored in a database or data structure. In other embodiments, when f is needed, the correspondence (4) identified in step 512 will be calculated first 'and then selectively stored to provide future reference 0. 装置 The device and method of the present invention include metadata. Components and corresponding components. The metadata element is associated with a video content identifier, such as a lion or a hash function, and the metadata element also corresponds to a space-time coordinate _, such as a (post) and video content identifier in a video sequence. In the space-time coordinates (x, y, t), χ corresponds to the spatial dimension, and t corresponds to the time dimension. The specific metadata element m ( 'x, y, z) is shown, where "να" represents the video content identifier. The να 'metadata in different faces may be configured with the forest_yuan data source. In the sum of money, the meta-data is the material of the order grade (for example, the metadata element m (VCI; x) = knowledge) and the time (four) is not detailed. 4 Under the condition of 'empty movie' 15 200951832 is not necessary. After the seat ^, and the corresponding relationship between the two different video content identifiers of the time and space, the function is expressed by the function (1) Wvcn, (10); xl, yl, tl), where (x2 y2, T2)^' and (A yi, ti) are space-time coordinates in the video sequence, and the video sequences vcn and VCI2 correspond. Video content from different sources can be mapped according to different kinds of enemy να (for example, YouTube video clips and CDs), and the video content and phase __ να can also be established differently for the same movie on the blue line, such as DVD. version. The general lookup associated with the video is presented in two phases (VCI〇; x〇 y〇, z〇). First, the new search has a corresponding να with v(10), where C (vao, VCI1; XO, y0, to), c (ν(10), VCI2; x〇 y〇 t〇),

c (VCIO, VaN; x0, y0, t0) is indicated. These correspondence conversion coordinates (X) are (xl, yl, t1), (x2, y2, t2), ..., (xN, yN, tN). Next, the system ◎ will fetch the metadata m(VCI1;xl,yl,tl), m(VCI2;x2,y2,t2),·..: (VON; χΝ, yN, tN). For special metadata components, the system will retrieve all the data or part of the it data. For example, if the metadata contains all the information about a movie, the system may only want to display the name and summary information of the movie' so the system will only extract some of the metadata. As discussed in the present invention, the associated generic video is searched to calculate the correspondence of the two video sequences. In some specific cases, the pair of time and space pairs has been calculated. In other cases, the correspondence of the time or the correspondence of the space is I. When the space is required, the embodiment of the universal video surface system Μ 16 200951832 first calculates the correspondence between the two video sequences at the time _ corresponding _ ϋ two identical video sequences. different

The correspondence between the two video phases can be a lot of methods. Because of the insertion and/or deletion in the - or two video sequences, the gap of the correspondence relationship will be caused. Therefore, the correspondence of the time of the video sequence is not Need to be a pair. A general description of the correspondence of time can be written as (u, (1), representing the time t1 in the first video sequence corresponding to the time β in the second video sequence. The programming algorithm, the Smith_Walt_n algorithm is searched for by the silk = the best calibration line between video sequences (eg, the correspondence of one or more gaps). The material used for calibration may be a video-based descriptor, as discussed in the present invention. DNA description symbol.) As described above _ 'The two video sequence touches _ can be used for two fixed view _ _ _ 贱 。. Typical _ sub, video content has a "original" version, This version is the most complete and authoritative video content. All other versions of the video content will be synced to the original version (for example, the correspondence is calculated using the original version). In a particular example The original version of the DVD version of the treasure of the original version of the treasure (the careful version and the version of BitTorrem) will be calibrated on the timeline of the DVD disc. Various systems and methods (4) identification, association Tracking, pairing, and adjusting video frames and video sequences. The specific embodiments of these different functions are discussed as follows: Video material includes time and space data, which contains two spatial dimensions and one time dimension (ie, two-dimensional Video images and time series of different video images.) We divide the correspondence between the corresponding positions of the two different age-age coffees and the space 200951832. The correspondence of time is expressed in the time granularity of the time between different video faces. (timegranularity): The video sequence is treated as an ordered-dimensional surface sequence, and the matching produces a correspondence between the pictures in the two sequences. The spatial correspondence is represented in the sub-surface granularity and found in the sequence. The corresponding pixel or domain between the two faces. The problem of the correspondence and similarity is closely related, and usually calculating a problem makes people infer that another problem is also being calculated. For example, we can define a large number of The _ _ part is similar, on the contrary, if _ has the same criteria for the similarity between different parts of the video sequence, I You can define the correspondence as the maximum range of the similar parts. Here we want to distinguish the similarities between the two types: semantic and visual. The "visual" similarity of objects means that they "seem like" For example, their pixel representations are similar. The "semantic" similarity means that the two objects represent the same concept. Semantic similarity is more broadly defined than visual similarity, for example, trucks and Ferrari is visually different, but semantically similar (both represent the concept of a car). Therefore, visual similarity is easier to quantify and evaluate, while semantic similarity is subjective and problem-related. Different angles, lighting conditions, editing, resolution, etc. cause the video signal to almost always have noise and distortion. This ideal similarity criterion should be no, and will change with some or other changes. In academic terms, if the criteria for similarity are similar to the description of the two objects, no matter how they are illuminated. In our case, the similarity does not change with the lighting conditions. The above described systems and methods allow matching video sequences to be unaffected by editing and distortion, and more precisely, these systems and methods provide visual similarities - 18 200951832 ^Temple-corresponding framework, which allows time Distortion (like frame rate change), time editing (removing or inserting face), spatial distortion (pixel value (fine_) different), and lang editing (deleting or adding content in the screen) will not change. In mathematics, the problem of spatio-temporal matching can be planned as: providing two video sequences, finding the coordinates (x, y, t) of the spatio-temporal system of the first-sequence t and the coordinates of the spatio-temporal system in the second sequence (x, , y', t,) correspondence. Think of video data as a three-dimensional array of pixels, and the matching problem of time and space can be

In order to find out the correspondence between two three-dimensional _, in general, this problem is complicated in the calculation of the juice (complexity is NP^ompiete), and such calculation is impractical. This is because without further simplification, the computing system will attempt to find a match for all possible subsets of pixels between the first and second sequences, which would be a fairly large number of operations. Stomach However, if the matching problem is divided into the following two separate programs, the matching problem can be greatly simplified: time matching and space matching. Among them, the problem of spatial matching is more complicated, because the video picture is two-dimensional, which will produce a large number of two-dimensional alignment of φ. On the contrary, although the one-dimensional time matching problem is still complicated, the video DNA or genomics dynamic-programming method discussed in the present invention can easily make one-dimensional (time) signals effectively match. "Fig. 6A" shows an example of the correspondence between the space of the video material and the correspondence of time. In the first phase 600 of Figure 6A, the matching of time is revealed (this step is discussed in more detail below). The matching of the time produces a correspondence between the coordinates "t" of the time in the subset of the first video sequence and the coordinates "t'" of the time in the subset of the second video sequence. By representing the matching of time, we need to avoid trying to match the two-dimensional space between all possible subsets of pixels in the video sequence (in essence, the three-dimensional matching problem). Moreover, this problem is reduced by size, so the matching of spaces must be represented between small subsets of the correspondences of the time of the partial video sequence. In other words, for spatial correspondence, large 3D matching problems are transformed into smaller 2D matching problems in relatively small portions of 2D video. For example, try to replace the "apple" string that matches the pixel "thing" in the entire video sequence of the upper layer with the thing corresponding to "apple" in all the video sequences in the lower layer. Thus, in "sequence A" and The most relevant small number of faces in "sequence b" have been examined. The following discussion is one of the typical short problems in video sequences. The correspondence of the time of such a video sequence is small, which can greatly reduce the problem of space matching. The spatial matching of the video data of the corresponding relationship in the second stage of the 6th 2' time in "Picture 6A" is revealed. The matching of the spaces produces a correspondence between the spatial coordinates (x, y) and (x', y') of the matching portion (eg, picture) of the time on the first and second video sequences ^ ' in the described system and In practice, pairing can become more intense, and it is constant for the distortion and editing of the content of the movie. In particular, the matching of time is stable for the short film sequence editing. Spatial pairing can make the film distortion and the editing of the film+sequence stable (for example, 'the different ratios of the rhymes' on the "Fig. 6A" and the different fruits' backgrounds). ▲ It is to be understood that the methods described here are normally at least included: processing 11 (usually on two computers with processors) and memory (passing 1 with megabits or On a computer system with a memory above giga bit, the existing processor suitable for performing this method is usually a computer that processes x86 N KiTPC n vuu, Power, ARM, and the like. Or be used as a shadow 20 200951832 : such as the processor, digital signal processor, and who is the processing benefits. The methods described here can be "C", "C++", "java", "system", "p_" in higher order 11 and operate on lower-order components or directly. The results of the research can be stored on memory storage media such as memory or flash drives: hard disk, cd, excitation, blue light and phase storage media. ❹ , see 贞 ( (video image) can be used - a small part of the special location feature points This video feature point is a point that can be found in the image to stabilize the η 2 "wide machine method. - A feature point will affect other feature points, such as 5 electric & Wei - sample 'typically bribe local shadows or feature points ^ week situation. The ship's often Wei Li to turn the image is rotated, modified to be presented later, and presented with different light sources. The feature point f is described as the vector information of the time-space side subset of the shadow image. For example, '- a point may be a 3-dimensional direction of a space-time edge, a direction color of a moving range 77 cloth', etc. basically, part of Feature points provide a description of an object body. All feature points construct a background. For example, an apple object on a computer advertisement and an apple object in a hot fruit have the same - part of the light spot describes the object, but the overall background is different. * For example, some feature points should include: . Like the corners described by c. Harris and M. Stephens at the 4th Alvey visi〇n Conference in the year of "A c〇mbined c〇mer — as you are." Harris comer detector and its variants. The SLFT theory described in D. G. Lowe, "Distinctive image features from 200951832 seale-invariantkeypoints", published in "International Journal of putef Vision" in 2004. • The motion vector obtained by decoding the video stream. • Guidance on the edge of time and space. • The distribution of colors. • Essential description. • Factor decomposition of pixels in known dictionaries, such as wavelets, curvelets, and so on. • Special objects in the prior art. Extending this idea to imagery, we can extract a video into a two-dimensional structure (two-dimensional space is formed by many 2D images, and one-dimensional space is formed by many video images). This 3D architecture can be used as a basic building block for presenting video sequences. As discussed earlier, thinking about video analytics in terms of biological nouns is quite helpful and inspired by biomorphology. Here, in this example, it is useful to think of feature points as atoms. The feature points condensed by multiple video frames are like nucleotides, and the video itself is like a sequence controlled by nucleotides. It's like a big DNA or RNA molecule. There are different interpretations of the spatial and temporal dimensions of the video sequence. The time dimension can be a command for a video material. We can say that a feature point is connected to the material. After the feature point. If we divide the video sequence into short intervals, we can think of it as a command-controlled sequence of video components. It contains a collection of feature points. As mentioned earlier, here we consider the sequence of a small portion of the nucleotides of the video material, and the signal we think a video is also like by $

Video submission of glycosides, called DNA DNA 〇 X 22 200951832 The analysis of the device and method proposed in this (4) can make the human sand not two-dimensional, two-dimensional and one-dimensional signals. Considering all the feature points, we get a three-dimensional (space-time) structure. Consider the time interval • and we can get a one-dimensional representation. Consider a 昼 in the sequence. We can transfer a two-dimensional representation. The same expression is used to carry out the pairing phase of space and space. The next step is a two-stage comparison method. In the first phase, the _ expression will be created when a video phase is reached. Each video phase will be divided into time intervals. The time interval here depends on 'not just a single video face' but a segment consisting of some video faces (for example, 3 to solid face). The time interval will be discussed in more depth here. For all time intervals, the actual video image will be extracted into a representation U. This is also called visual nude (tide). In the expression, the key feature points of the interval of packet 3 are discarded. The spatiotemporal coordinates of the feature points extract and compress this sequence of feature points. In other words, we only need to start calculating the non-state of the feature point. In a nutshell, this seems to record the number of descriptors O (descriptor) and the number of descriptive symbols of different feature point patterns. Each time-divided video signal (which we will call "(4) acid", which is similar to a biochemical acid) will be represented as an unsorted collection or a feature. The "bag" of points (or Is a bag of characterization symbols). Therefore, if each feature point is thought of as a "visual atom," then the "feature packet" is a time segment of a particular video, which can be called a "nucleotide." The representation of the time segments of the different videos is then arranged into a sequence of "sequences" or maps (calendar) (video DNA). In the present invention, we usually use the term "nucleotide" instead of the "characteristic package" because it helps to guide the square DNA of β by a useful information method that approximates the video analysis program. The _sequence' can be used in a similar way. For example, in the analysis of the alignment and calibration dna, an important problem is to find the correspondence between the maximum similarity and the smallest gap of the state ❹ = _: The algorithm of the system method of := t, the _ _ back of the class _ 资 资, = algorithm can be used to align two different video signals. After the additional =: phase is compared with the two video media, the corresponding relationship between the time and the correlation can be found. In the first video, I have two more; Daisy" (pixel group, which can be compared in the second video. Inside 6 = ship can be Wton two (four) should be seen in the image of the valley between the valleys Correspondence of space. - Back to the second stage after the pan. 'We will not discard the space-time coordinates of the feature points. In Ο -IW and 'mother facets are represented as a two-dimensional feature point structure, and we The coordinates of the feature points are preserved. For the purpose of the second phase, the spatial matching of 2, the comparison of the age, the more standard feature point algorithm, and the computer vision file used by the first are used. Object recognition and other flipping, object-based analysis is necessary _ (vldeGge_ies)" method provides more meaningful than the previous technology. Money, the secret of the hair riding and the green provide a = 彳 description symbol for the high identification This discrepancy is not only the description of the object, but also the time series _News, like the sequence of these bribes. The sequence of 24,518,518. Although some existing methods say that the best ability to identify is through a larger Digital precision The feature points were obtained, but we found that this is not the case. Unexpectedly, when the system and method described in the present invention are compared with the prior art, we find that the result is time support (like chronological order). In the group of most special points, it is more discriminative power than using a very large number of different descriptive symbols. For example, the accuracy of the object description is usually straightforward. The prior art is " The brute f〇rce method to increase the precision of the sensibility will touch more feature points and feature descriptors, but each feature point and feature description symbol is generated by the intensive operation. The "Violence Law" will quickly reach the decreasing position due to the high computational cost. Innocent, we have found a way to increase the accuracy of the description of the object, and other aspects must first be based on two strengths. Sequence (increasing the computational load according to the order of the two nearby strengths) to increase the visual vocabulary size, so that the device according to the present invention can be used The method uses a low-intensity calculation for simple comparison. The apparatus and method described in the present invention are used to improve the accuracy, and we avoid increasing the number of features ❹=lying, and the analysis of the domain-to-office resolution to increase the accuracy. This will be done by simply adding two more "nucleic acid" (like using a smaller time segmentation in a video analysis) to a "video DNA" sequence comparison. The strength of the σ is particularly difficult to rely on, the hairdressing method of the hair ride can achieve the South precision 'and will be more efficient from the point of view of the calculation. J technology method 'like j Sivic and a zisserman's "Video Google, a Text retrieval approach to object matching in video j is a collection of images, in which a large number of feature points "vocabulary" (more than one million elements) are used in order to obtain high recognition power. In contrast, 25 200951832 use time support can be less (10) vocabulary (get the same or better results, and increase the efficiency of the calculation. s read) The first advantage is based on content Back to the application, the method and method permit the fetching of objects of interest and content disparity. The time series is set to describe the additional information of the object, as well as the description of the object itself. , β shows the same object (Apple 61〇) in two different orders in the “Picture 6”: fruit 612 and computer 614. In the first case, ^ 働 働 纽 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新 新In the second case, the "Apple" object is in the middle of the fish pen and the computer, which gives the object a computer meaning. Here, the system and method are complex enough to distinguish the differences between these contexts. Therefore, the presentation of the video map/visualist DNA in both cases will not be the case, despite the fact that the object itself is the same. In contrast, prior art methods such as Sivic and Zis_an did not take into account the prophetic consequences of video content, and thus could not distinguish between the two different objects "Apple j in the above example. The second advantage is the description The "video gene" approach allows partial comparisons and pairing of video sequences to be performed in many different ways. Just as the method in bioinformatics allows different DNA sequences to be compared, two different video DNA sequences can be paired, although there are some different video frames (nucleotides), such as insertions or gaps. This is especially important when invariant video modifications (such as time editing) are needed. For example, the video DNA of the movie and the version of the inserted ad need to be correctly paired. Figure 7 shows a conceptual outline of the video mapping/video 26 200951832 DNA representation of the generated video sequence. The program contains the following stages. In the first phase, 7〇2, a local feature detector is used to detect the interest points of the video sequence Yin. Suitable Features The detector includes a Corner # comer detector ("Harbin # comer detector", published in C. Harris and M. Stephens, "A combined comer and edge detector", 1981, or by D. G. · Lowe's 2004 feature-based feature detector based on scale-independent feature point conversion (SIFT) scale space published in ucv's Distinctive image features from scale-invariant keypoints. 〇 Points of interest can be traced to several video frames to fix points that are meaningless or inconsistent in time (eg, appearing in a short period of time). This section will be described in more detail later. Then the remaining points are using local characterization symbols, for example: the scale is not (4), the point is (SIFT) is in the direction of the local distribution gradient, or by H·Bay, T. Tuytelaars, and L·van G〇〇1 at 2〇〇6 The accelerated robust feature points disclosed in the published mbuStfeatures (sur^ algorithm. The feature points and description algorithms should be designed to be robust or invariant to spatially distorted video sequences (eg, resolution changes) , noise compression etc. ©, the position of the feature point and the corresponding feature description symbol form the representation level of the most base video sequence. (4) Stage W 'video phase is divided into handsome_interval 7G6, which is often two (usually 3 to 30 ).) Such a segmentation can be phonetic...+1/' can be based on feature points tracked from previous stages. It is worthwhile to modify the video of the ΓΓΓ video, such as the space-time position of the first (4) (Feature point coordinates) are not used at this stage. 27 200951832 Conversely, the information in the 'time interval' is described as a method of using the "feature package". Here, 'similar to Sivic and Zis_an's proposed Method, all characterization symbols represent the use of visual vocabulary (acquired from the aggregation of representative descriptive symbols, for example, by vectorized green). Each transcript is visually touched. The corresponding closest element Lai. As mentioned above, the point H in this way is also a visual Xuan. According to this push, the thief can be regarded as a visual element of the "periodic table of elements". Unlike the prior art methods proposed by Sivic and Zisserman, here we discard those that are referred to as "representatives", "visual nuclear _", "nuclear acid", or occasionally referred to as "feature packs" 710. The spatial coordinates of the feature points, and the frequency of occurrence of different visual atoms in the time interval as a histogram (group or inward). Here the "visual nucleotides" 712 representing a certain number of video pictures from the video are basically The above is based on the discarding space coordinates and simply counting the frequency of occurrence of the "feature package" (this process is regarded as "package function" or "group function"). If the visual element is marked A collection is used to describe the content of each "feature package", then a visual nuclear thief can represent a histogram or sparse vector of the silk form. For the sake of the paste, if the "feature package" describes several video images, it contains There are three cases of the first feature point, two cases including the second feature point, and none of the cases are examples of the third feature point, which is used to describe the visual nucleotide of the video image or It can be represented by a histogram or a vector (3, 2, 〇). In this example, the visual nucleus (321) is represented by a histogram or vector (〇, 〇, 〇, 4, 〇, 〇〇) 〇,〇,5,0). ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 28 200951832 feature points (a part of the old feature points will belong to the original video and another part of the new feature points will correspond to the overlay layer). If the size of the overlap is not very obvious (in other words, most of the information in the facet belongs to the original video), you only need to have a certain proportion of feature points in the "feature package" (such as the evacuation vector). Can # correctly match two visual nucleotides. Finally, all of the visual nucleotides (or signatures) are aggregated into a sequence of sequences in the video map or video DNA*714. Each image (or visual nucleus®, “package”, histogram, or sparsevector) can be thought of as an inferred letter in an infinite alphabet, therefore, video DNA A sequence of words in a broad sense. The time matching of the two video sequences can match the corresponding video DNA using a plurality of different algorithms. In order to match the DNA sequence of the organism, this can be extended from a very simple "matching/unmatched algorithm" to a "d matrix" algorithm suitable for biological information to similar precisions that are used in biological information. Algorithm. Some of the more complex bioinformatics algorithms, such as those proposed by S. BNeedleman and C. D Wunsch in "A general method applicable to the search for similarities in the amino acid sequence of two proteins" in 1970 • Needleman- The Wunsch algorithm, the Smith-Waterman algorithm proposed by TF Smith and MS * Waterman in "Identification of common molecular subsequences" in 1981, and the "Basic Local Alignment Search Tool" by SF Alschul et al. Heuristic algorithms such as Basic Local Alignment Search Tool (BLAST). In general, a suitable sequence matching algorithm will operate based on the score (or distance) that defines the match and the quality of the match between the two video sequences. The matching score contains 29 200951832 Two main readings: the similarity (or distance) between nuclear hearing and gappenalty, and the standard algorithm for how to try to split the sequence without introducing a sequence. In order to do this, the distance between the nuclear bud acid in the first video and the corresponding (four) acid in the second video must be checked by Wei Xue's scale. How to change the "feature package" from the sequence of the first video sequence to the "special package" in the second video sequence? Similar values can be measured by measuring the matrix of two thieves with similar or dissimilar patterns. In the simple _ shawl, the similar filament _Euclideandistance or correlation vector (feature pack) represents each nucleotide. If you want to allow partial similarity (which happens often, especially if the visual kernel_may contain different feature points when editing in space), you should use a more complex matrix of weights or rejection of outliers. More complex distances may also take into account the probability of a change between the two cores: if two different cores are mutually different, the two different cores have similar possibilities. For example, consider a first video image of the first video sequence, a second video image of the same first video sequence, and a video overiay. Obviously, many of the video feature points in the package described in the first video will be similar to the video feature points in the package described in the second video, and because the video overlap '"variation" is different here. Video feature points. Void penalty is a function of introducing a gap between sequence nucleotides. If you use a linear penalty, simply find the gap multiplied by the number of preset constants. More complex gap penalties may take into account the probability that a gap will occur, for example, based on the statistical distribution of the position and length of the ad in the content. The following discussion identifies cases of similarity and dissimilarity between biological DNA and video dNA. Since the systems and methods discussed herein essentially convert the problem of matching the corresponding portion of the different media 30 200951832 into a problem that allows for similarity to the problem with the paired bio-dna sequence, this analogy can be examined more closely. And get some deeper understanding. Because the pairing technique of DNA sequences is carried out in a relatively late stage of development, which is related to the matching technology of video, the system and method have unexpected results, which shows some post-stage DNA bioinformatics research methods. How can the technology be unexpectedly applied to different areas of paired video signals. As discussed earlier, at the conceptual level, there is a strong similarity between the structure of biological DNA and the method of describing video DNA. A biological DNA is a sequence consisting of nucleotides, and video DNA is composed of visual nucleotides (a signature package of multiple video screens). A nuclear bitter acid in a living organism is composed of a group of elements in the periodic table of the elements: in the same way, the visual (four) acid is discarded by visual words (a standardized packing area with ten different feature points). A set of feature sets consisting of visual atoms (that is, feature points). ❹ Figure 8 is a graphical representation of the extracted video signal 8(9) and the DNA molecules and their constituents (the analogy between the nuclear phytic acid and the atomic structure, "monthly "video DNA" name peak_^ There is a conceptual similarity, and there is a slight difference in the appearance of coffee A. First, it appears in the second, the scale of the β, f sub-periodic periodic table is small, usually only include some elements (such as anti-oxygen, phosphorus, Nitrogen, etc.). The size of the vocabulary in N is typically at least -i side C atom. (44mw^〇S has several thousand to several million visual elements, biting hundreds of typical nuclear buds. The number of atoms in the acid molecule is relatively small (the number of characteristic capsules). The "visual atom" (feature point) is worth a thousand dollars. When it comes to biological acid wipes, the space is closed to 200951832 and the relationship between atoms It is important that in a visual nucleotide, this relationship between feature points (that is, the reconciliation between feature points) is usually not emphasized or ignored. The third 'biological DNA sequence in different nucleosides The amount of acid is small—usually four in the DNA sequence (rA) rT", rG", rc") nuclear acid, twenty in the protein sequence. In video DNA, each video nucleotide usually contains at least a few hundred or a few A "feature package" of thousands of different feature points can be represented by a histogram or a vector. Therefore, if a group or a packed area, such as 500 or 1000 standard feature points, is regarded as a standard, frequency analysis When selected, the “features” will be a coefficient of 500 or 1000 standard feature points appearing in a series of video frames described as “nucleotides” or “features”. Multiples consist of histograms or vectors, so the number of permutations in this amount that may individually represent different video nucleotides is large. These de facto differences will make the matching of video DNA only similar in nature to the creature. Pairing of sequences. In some ways, the problem of video pairing is more difficult, while in other respects it is simpler. More specifically, the pairing calculus is different in the face-to-face orientation. First, in Health In the sequence, since the number of different nucleotides is small, the pairing—the scores of the two nuclear collisions can be represented by a simple “pairing” or “not able to satisfactorily” result. That is, the nucleotides of one organism may be Is "A", "T", "G" or "C, and it is necessary to distinguish whether there is a pair of rA" and rA", or no. In contrast, the nickname of each video DN wiper It is a series of histograms, vectors or "features" of thousands of different coefficients. It is complicated to operate. Therefore, for video DNA, we need to use a more kind of core 32 200951832 The concept of "score method" or "distance method" between glycosides. Such a score can be regarded as a distance method between a histogram or a vector. In other words, how far is the distance between any two different “features”? ‘On the other hand, many different concepts, such as homology scores, embedding, deletions, r-point mutations, and other similar concepts, have significant similarities between these two different domains. "Picture 9" presents, in one embodiment, a © calculated video DNA in an input video sequence. The video DNA calculation process will receive 9 视频 video data and include the following stages: Feature Detection 1000, Feature Description 2, Feature Trimming 3000, Feature Reproduction 4000, Time Interval Segmentation 5000, and Vision Atom The gathering of 6000. The output of the process is a video DNA. Some stages may be performed in different embodiments or not at all. The following description details the different embodiments of the above video DNA calculation stages. As shown in the figure, the video sequence is divided into intervals of a set of time. "Fig." indicates that in one embodiment, the length of the video time interval 1〇2〇 is fixed (e.g., '1 second) and does not overlap. In another embodiment, time interval 1022 has some overlapping portions. Here, each view_ may consist of as many video faces as there are in i 'seconds (or one group to be), depending on the ratio of frames per second 'may be 10, 16 Faces of 24, 30, 60 or some minor combinations. In another embodiment, 'these intervals are changed according to the position of the lens (scene), or suddenly under the premise of satisfying the two consecutive pictures (reference brain > 14 may be miscellaneous) The next lens of the larvae is under the # every structure, these new path numbers will be calculated and will replace the path number of the original 33 200951832. If the missing path number exceeds certain limits, then replace the original road (4) The hearing will also exceed the _-face, this picture is reduced to a lens. If the lens has been in some places, in a video of this video, the nuclear acid may be from hundreds or thousands of videos. If the picture is very long, in another embodiment, _ is continuous and synchronized in each class (Ref. 1026) ❹ Feature Test 1〇〇〇 (Refer to Figure 9) A feature detector operates on the video data_ and produces a set of N invariant feature points, {(from, fat (please refer to the "9th" face), where x, y, t represents the space-time correspondence of feature points, respectively. Feature detection step 1000 has been shown in detail in "U-figure" and there is an example of using green. The side of the machine is executed under the screen of -Jiang. At time t, a facet, a set of feature points of ^ is tilted. A typical feature point has a two-dimensional edge or secret. The standard calculus of the invariant feature point _ can be considered in _!<. For example, 秘 ❻ ❻ track, tracker (KLT)...etc.

The representative values of Nt range from 1〇 to 面. In a particular embodiment,

The value of Nt is 1GG, 2GG, ..., face. In another embodiment, the value of the valence is the result of the pre-set _' and is a feature _ algorithm. In another embodiment, the feature _ is executed in a data that exists in _ and space, and produces a 彳 σ 丨 (the feature lie algorithm of wn can be used for this purpose. , 10, and 2000 (please refer to "Section 9"). Each feature point in the debt test "when" a feature description symbol will be calculated, resulting in: a collection of symbols ("9th figure The path is in accordance with the feature point. A temple: 34 200951832 Γίΐ is a representative of the local visual teacher. In the computer world, there are things like = SIFT or petal feature descriptors. It is expressed as a nail-empty _ vector, a feature description symbol F=128, and a feature point symbol F=64.] If the implementation is _, the ship description is in the r In the case of the calculations, that is to say they represent the pixels on the screen of a certain 曰, 曰目=. The standard characterization symbols are as detailed in SIFT or SURF. In another embodiment, the characterization Symbol is present

Branches, Lang, and New, they represent pixels that exist in time and space. The characteristics of the bottle are described as a bottle to be used on it. J is 3000 (refer to "Figure 9"). Here, at all feature points, find a subset of the feature points 3_. In the case of the non-like embodiment, the consistency may be as dark as possible—the miscellaneous features (the feature points are not sudden (10) and the positions at the adjacent time are similar), and are temporarily consistent (the feature points are not present or suddenly) The ground disappears, or the consistency of time and space (the combination mentioned above). In the example of the ceremony, in the "12th picture", in order to find consistent features, the point: execution of the memorial. In the feature point tracking algorithm 3100, it tries to find out that several and two places are presented in a feature that has a sufficiently large and connected sorting picture, so that it can be transferred in a picture. · False feature points. The k-fake feature points interfere with the image of the county. For example, removing the feature points from the mirror will improve the accuracy and visual content in the face. 35 200951832's chase: Taste: 5:: 里' Use - a screen based tracking method. In this way: the person goes to find the corresponding feature of the two features in the face, and usually the addition of the teacher (four); In =r:, multiple screens are tracked in the same time. Follow-up state 3100 output trajectory τ 2 table _ start and end of time, and H is just the time of lang. There are some points to determine these key points (10) - the special features in the recordings are changed here) ^ phase, movement (the position of the feature points does not change significantly with the trajectory), or the cypress It is the money to lose (4) the ship's point of recourse can be converted, the inspection of the lion-mixed and recorded tracks are deleted. In an embodiment, the specific nails are removed, and some of these tracks will be removed due to non-compliance. In another embodiment, the manifestation of a highly variable (sudden movement) showing a spatial similarity will be deleted. In another embodiment, the feature descriptors of those features, if their profiling also shows a high probability of variability, then k2 will also be removed. The result of the removal is _ a subset m of the trajectory τ. In one of the embodiments, the feature set 仏, read ^ and the corresponding descriptor will be calculated in the time of -wei t _, and the chase will be initialized to the ' Kalman filter (fllter) will be predicted in the next 昼The area is wide, and the position of the feature point in the middle is (10). This collection ^ heart light (four) should be in the descriptor ί /; 1 = 被 is calculated under the picture. Each feature point ~^, 乂 乂 都 36 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 Corresponds to the closest descriptor selected. When the face is adjacent to the sort • No proper pairing occurs, the track is terminated (failed). Only those track points that meet a sufficiently short time are saved. In one embodiment In the Kalman filter, the constant rate model (vel〇dty) is used together, and the estimated feature point position covariance is used to determine the search radius of the next picture. 0 A feature point based on tracking ("Fig. 12" step 3200) The trimmed embodiment is described in more detail in "Figure 13." The input feature point position 1010, the corresponding feature description symbol 2010, and the feature point's execution 3丨丨〇, the persecution in each track, degree The "d", the movement difference "mv", and the description symbol variation "do" will be calculated. These values will pass the - (4) coffee and decision-making rules, and will remove the trajectory that is too long or too variant. The result is a subset of the feature points that have been traced after the trimming. A possible decision rule for discarding the trajectory is as follows: ❹(d>th_d) AND (mv &lt; th_mv) AND (dv &lt; th—dv) where th_d is the duration of the gate, "th-mv" is the gate of the mobile mutation, and "th_dv" is the threshold for describing the symbolic variation. 'Character representation 4000: Back to Fig. 9 shows that the step side expresses the feature points on the trimmed track in visual vocabulary. The result of this stage produces a series of visual atomic sides. The visual vocabulary is the aggregation (visual element) of the κ-like overview symbol. It is represented here. The visual vocabulary can be pre-computed, ^ column says 'a large number of feature points are collected from the set of table read frequency phases and represent the clarification on the descriptive symbols. In different embodiments, the value You can substitute 1 side, 2〇〇〇, 37 200951832 3000,..., loooooo. &gt; Replace each feature point 兀· according to the number of pixels/the visual capsule with the descriptive symbol closest to the feature point/. In the implementation of the fiscal, to the closest calculation (nearest ^eighb^r algorithm) to find the representation of the feature point z•, where * is the reference for identifying the symbol space. In another embodiment, the approximate proximity algorithm (approximatenearestneighborh〇〇d) is used. Since feature points / are represented as, they are called visual atoms. - In an embodiment, before the feature points in the visual vocabulary are represented, $ 特征 will be found for each feature's feature point representation. The trajectory obtains the mean, median or _ of the characterization symbols from the descriptive symbols of the feature points. In a solid yoke, the feature points that are not discernible will be trimmed. Feature points that are not discerning 特征 those feature points that are almost at the same distance from most visual atoms. These feature points can be judged from the ratio of the distance from the first closest neighbor to the second nearest neighbor. Visual Atomic Aggregation 6000: For each time interval calculated in Step 5, the visual atoms are aggregated into visual nucleotides. The visual nucleotide sequence (video ❹ DNA6010) was output. The visual nucleotide j is established as a histogram of the container (κ is the size of the visual vocabulary), and the nth container is the number of visual edges of the η type appearing in the time interval. In one embodiment, the histogram in the interval [Wj adds weight to the temporal position of the visual atom in the interval according to the following formula: Κ = i:lj =n where ' w (〇 is a weighting function, ~ is the value of the nth container in the histogram. In one embodiment, the weight is set to the maximum value at the center of the interval and decreases to the edge 38 200951832 of the interval 'for example, according to the Gaussian formula w (0 = exp f2 1 • In another embodiment, 'detecting the interval [with clipping of the lens within yj and from the boundary of the lens to the center of the interval) is set to zero. In a particular embodiment, the container in the histogram is more Weighting reduces the impact of unreliable containers. For example, the weight of the nth container is inversely proportional to the typical frequency of the visual atom of type η. This type of weight is like the weight of the file frequency in the text search engine 0. In another embodiment, the weight of the nth container is inversely proportional to the variation of the nth container calculated from the representative of the typical mutation, and is proportional to the variation of the nth trough of the same content. The two video sequences compute video DNA that can be matched (corrected) in time according to the description below. In one embodiment, the serialization of the visual nucleotides represented by the video DNA is queried in the database. The sequence of the visual nucleotide represented by the video DNA in the {ίγ=ι is calculated in the following manner. The match between the two sequences, the nucleotide 4 corresponds to the nucleotide ~, or Corresponding to the gap between the nucleotides \ and ~, the same, the nucleotide ~ corresponding to the nucleotide A, or the corresponding gap between the core and the A+1. {ςτ,.Α and the match between It is possible to have a sequence of fe corresponding to K, a person 丨 [=1, a sequence having G gaps, and a sequence having a G' gap, {(4, 、), where (^, Nine) indicates the length of the gap between the nucleotide gas and L+1/», and the subsequence indicates the similarity '仏, human 乂) indicates the length of the gap between nucleotides A and 氕+1/, The subsequences represent similarities. The match will be configured according to the following formula: 39 200951832 GG* S = Σσ(^Λ ) + + 'ZsiLJnJ») 众=1 w=*ln Article 1 where 〇(%, &amp;) is a scoring equation for calculating the degree of similarity between nucleotide a and nucleotide ~, which is a gap Penalty. According to the above description, many alternative algorithms are used to make pairing calculations, including simple to very complex. In one embodiment of the invention, the Needleman-Wunsch algorithm is used to The maximum total score is $ to search for pairing, in another embodiment the Smith-Waterman algorithm is used, and in another embodiment, the BLAST algorithm is used. In the freely chosen embodiment, the paired maximum total score S is utilized The following methods are done by ®. In the first phase, a good pairing with a small fixed length w is found between the query and the sort in the database. These good pairs are called seeds. In the first and second 匕S, the method extends the pairing from the two directions of the seed. An un-gapped alignment process attempts to improve the alignment score using an initial seed pair that extends length w in each direction. Inserts and deletes are not considered at this stage. If you find a gap-free alignment of the high scores, the database sequence jumps directly to the third stage. In the third stage, the query order ship #料料_ ❹ gaped aiignment can be obtained by the Smith_Waterman algorithm. * In an embodiment of the invention, the gap penalty is linear and can be represented by = &lt;, where a is a parameter. In other embodiments, the gap penalty is similar to that of the machine, nine people = 0 + 4]), where 々 is another parameter. In the examples, the scoring equation kisses, Λ) describes a histogram representing the nucleotide * * and the _ similarity of the histogram representing the nucleus. In another embodiment, 200951832 similarity can be calculated using the inner product of <^>. In a freely chosen embodiment, the similarity can maximize the recognition power of the scoring equation by weighting a vector weight derived from the training material. Alternatively, the scoring equation #% ^ ) is the inverse of the distance between the histogram A of the nucleotide % and the histogram &amp; In other • embodiments, the distance can be calculated from the Lp reference (Norm) 聃 - . In a particular embodiment, the distance is the Kullback_Leiber split between the histograms. In other embodiments, the distance is the EarthMover distance between the histograms. In a particular implementation, the scoring equation σ ") is proportional to the probability that the nuclear thief ~ mutates into another nucleus or space distortion applied to the basic ((10) touchdown) video sequence. That is to say, the lens (tetra) acid (10) histogram is mutated into another probability representation of the histogram/Γ of the nucleotide gas. In this example, the probability can be estimated by P{hW) = Y\P{hn\hn) η 'When MW is the histogram (4) the probability that the value of the second bin (bin) changes to . The probability paper can be obtained from the actual training data of each container in the actual experience. In another example, Bayes theory is used to represent the scoring equation as the machine p{h'\h)=^)pm m where p_ is calculated using the previously explained method. Household ρ^)-ψΛΚ) (1) The table is not 41 200951832 where 'w is the probability of measuring _ in the histogram A and independent for each container from the actual experience two 乂, this usually has _, not only - Second reading practice. Between different videos, and to find the first two-body face or time series from two-image group of pixels, the second space alignment in the video is not allowed to '' The second one is used to compare different sources or resolutions to ° or 'this is sometimes a flaw. Shooting a TV screen may want to be accurate: two = what. Among the two examples, it is useful to determine the spatial alignment and time (picture number) pairs in the embodiment of the present invention that determines two different videos, representing the auditory, and representative databases. Sequence Timing = Medium_Zone (10) Vision: "The missing time interval 匕, 纥] is the best pairing. The spatial correlation between 7 can be calculated by the following method: Descriptive from. Another - 昼 - can be composed of two sets of two r to the two of the two descriptors. The pairing at the corresponding location is rejected. The correlation score (c_spondingp〇int) can be expressed as ^y〇,{x'Jk,y.j 〇 ‘this can be found by the minimum value

In the JU algorithm implementation, the random sample consensus (^NSAC) can be minimized. In another embodiment, the minimum is used to re-weight the least squares 42 200951832 Θ Uemtivdyeweighted least squares fming) The algorithm performs a rotation 'size, or a twisting conversion is often useful ^ in one of the embodiments, the conversion The type of Τ is r cos^ sin^ ιι\ * t Τ -sin^ cos0 0 0 In another embodiment, the type of the conversion formula T is cos^ sin0 u\ T_. αύηθ crcos0 ν 0 0 1 Ο in other In one embodiment, the type of the conversion Τ is Τ bcd ν 0 0 1 :ive In other embodiments, the pattern of the conversion τ is a projection transformation (Pr0 fine transformation). Finding the space-time correspondence between two sequences can be represented by "Fig. 14". The program contains the following stages.

1. Video DNA calculation: Two video data sets 9〇〇 and 9〇1 are input to a video DNA calculation stage 141〇. Stage 1410 will be shown in detail in steps 1 〇〇〇, 2 〇〇〇, 3 〇〇〇, 4000 of "Fig. 9". This phase can be performed online or calculated and stored. 2. Time alignment: The resulting video DNAs 6010 and 6〇11 are input to a time alignment phase 1420' to calculate the time correspondence. 1425. The time-correspondence relationship is essentially a transformation that shifts from the video coordinate system of the video data to the video material 901. 3. Spatial alignment: The spatial correspondence 1425 is used in stage 1430 to select a subset of time correspondences for the video 2009 200951832 frequency 900 and video 901. The selected subsets 1435 and 1436 of video 900 and video 901 are input to spatial alignment stage 1420, respectively, to compute spatial correspondence 1445. The spatial correspondence is essentially a transformation from the spatial coordinate system of video material 900 to video and data 901. An embodiment discusses the following in which the video DNA video sequence of the input video sequence is calculated as shown in Fig. 9. The video DNA is input into the video material 900 during the calculation process, and includes the following stages: feature detection excitation, feature description 2〇〇〇, feature trimming 3000, feature representation 4000, time interval segmentation 50〇〇, and visual atomic clustering 6000 . The program outputs a video DNA of 6〇1〇. Feature Detection 1000: - A SURF Feature Detector (described in r Speeded Up (10) Features", from the 9th May of the 9th Computer Vision Europe Symposium) in each of the video sequences 900 The screen operates independently, and each of the "faces" produces a Nt=150 strongest feature point position of 1〇1〇 (please refer to "9th figure"). Feature Description 2000: For each feature point in the feature extraction phase, a 64-dimensional SURF feature descriptor is calculated, just like the 9th Computer Vision Europe Symposium in May 2002. An article "Speeded

Up Robust *

The same as described in Features. Feature Trimming 3000: This is an optional step and has not been implemented in this embodiment. The &amp; feature indicates 4000. The feature points represent visual vocabulary consisting of κ=1〇〇〇 entries. The representative element is calculated using a similar nearest neighbor algorithm (叩近^ her nearest neighbor algorithm), published in 1993, "Acm_siam 44 200951832

Symposium on Discrete Algorithms" (SODA), Issue 271 to 280, ‘S.Aiya and D. De Mount, mentioned “Approximate Nearest Neighbor

Description of Searching. Only feature points to the second nearest neighbor that are below 9〇% of the nearest neighbor are retained. The result of this phase is a collection of visual atoms of 4〇1〇. The feature point description phase of the visual vocabulary is obtained by pre-calculating the content of a set of classified images from a sequence of seven hundred and fifty thousand feature descriptors as a training resource. The K-means algorithm is used to divide the quantitative training cluster into 1000 clusters. In order to reduce the computational burden, the nearest neighbor search in the K-means algorithm is replaced by its most approximate variable, which was also published in the third issue of "^. ACM-SIAM Symposium on Discrete Algorithms" (SODA) in 1993. From pages 271 to 280, "Approximate Nearest Neighbor Searching" by S. Arya and DM Mount has a description. Time interval segmentation 5000: The video sequence is divided into a set of time intervals 1020 of a fixed length of one second (refer to "Number" 1〇图”).视觉 Vision Atomic Aggregation 6000: For each time zone that is calculated at stage 5000, the visual nucleotides are aggregated inside the visual atom. The order * column (video DNA 6010) produced by the visual nucleotides is the output in the process. The visual nucleotides are created to image a histogram of 1000 bins, and the nth container is responsible for counting the number of time intervals in which the visual atom appears in the form η. After two different or more different video DNAs are produced, the video DNA generated from these materials can be checked for their correspondence and pairing results in the following ways: Pairing of time 1420 (please refer to Figure 14) Using the parameters of the SWAT algorithm and the linear gap penalty " = 5 and # = 3, the weighted scoring equation is used. 45 200951832

a(h,h') = Weights can be calculated based on past experience. For this purpose, various training video sequences can be transformed into random-space and temporal distortions, including changes in blur, resolution, aspect ratio, and face velocity, and their video ships can be calculated. The peri- (iv) visual (iv) acid suspension is the same as the visual nucleotide difference corresponding to each container estimated from the deformation. For each container n, the weight 4 is set to the difference between the latter two.

Space matching 1440 (please refer to "Fig. 14"): The corrected space can be completed by the feature points being two 1 second corresponding segments representing the video data of the two sets _ and 9〇1, which is from the previous (d) Inter-correction phase coffee. - Each feature point in the face, in other areas _ mixed with the Euclid distance to minimize their respective. The oversimplified output is a collection of two m should be characteristic points {(Ά,, )}, {〇~,〆,,,;)}

Once the corresponding location is found, the RANSAC algorithm can be used to perform the conversion of the form f a b u T= -6 c ν Ο 1&gt; for the corresponding set. In another aspect, the invention is a space-time paired digital video material comprising a plurality of temporal (four) video frames. In this fiber, the method includes a step that is difficult to match when performing digital video data. [The digital image is included in most of the time matching video frames to obtain a similar matrix, and the spatial matching of the similar matrix represents the use of each video frame. The representativeness of the matching score, a her component, a gap 46 200951832 penalty component, and a representation that operates using a local correction algorithm (like a genome-based pairing algorithm, or other suitable algorithm) And the implementation of spatial matching digital video data, including a variety of video face time matching using the obtained similar matrix. The step of spatial matching in this is essentially independent of the step of performing the time-matching. The above method can also be applied to the Needleman-Wunsch algorithm, SWAT or other similar secret algorithm. The above method can be implemented using a genomic pairing algorithm, such as a basic local correction search tool that can be used to compare biological sequences, protein or nucleotide DNA sorting algorithms. The above method may further comprise performing local feature detection on the digital video material, wherein the digital video material comprises a majority of the time-matched video images to detect the customer's interest points; and the digits are segmented by the interest points, where the digits are digitized The frequency data includes a majority of time-matched video frames to a majority of time intervals, where performing spatial pairing and time pairing operates over a number of time intervals. On the other hand, this method may determine the temporal and spatial correspondence of the video material, and include several steps, such as inputting video data; representing the video data with an ordered sequence of visual nucleotides; The visual nucleotide determines a subset of the time corresponding to the video data; calculates the correspondence between the spatial and temporal sub-sets in the video data; and outputs the space-time correspondence between the video data subsets. Type of input data: In this regard, the video material may be a collection of image sequences, a query that may be a video material, and a video data set, possibly a subset of a single video sequence, or a modified video sequence in a collection of video material. a subset of. Furthermore, the spatio-temporal correspondence is established by querying at least a subset of the video sequences in the video material and at least a subset of the video sequences in the video data set. In a particular embodiment, the spatiotemporal correspondence can be established between a subset of the video sequences of the query video material and a subset of the video sequences of the visual woven set. The money mentioned in the above-mentioned "Lai (4) material" contains a modified subset of the video data set, and the proposed modification consists of the following one or more items: - • changing the picture rate; • changing the spatial resolution Degrees; • constituting an uneven space; • modifying histograms; • overwriting new video content; • increasing the time of new video content. , Nucleic acid cleavage: In the other-hybridization, the method and method include the video data of the interval between the divided trees, and the visual nuclear acid can be calculated for each ship. Interval Period: In another variation, the system and method of the present invention are capable of segmenting video material into a time interval of a fixed period or a time period of an indefinite period. The start and end time of the time zone can be calculated based on the lens shift of the fabric towel. It is also capable of recording time intervals that can be partially overlapping or not partially overlapping. , visual collision calculation: In another change, the visual kernel _ (as mentioned above, · = to describe the visual data of the time interval in the visual content) can also be calculated in the following steps. • Visual data representing a time interval As a collection of visual atoms; • Construct nucleotides with at least one visual atom function. 々 In this calculation, the function may be a feature point (visual atom) histogram of the frequency appearing in the time interval, or the function may be a kind of appearance in the time interval. If it is 1 can be attributed to the combination of Wei Xuan can be as follows: Set square + shape plus • Time drop of visual atom in time interval; ' • Spatial drop of visual atom in time interval; * • Effective value of visual atom . Different feature points or visual originality on the nucleus acid. · In the implementation, the weighted interval (for example, \· = 2 = ❹ all feature points are treated equally. For example, in an alternative weighting The weight of the square οβ 々 tea can be the maximum weighting of the rain function in the interval. The weighting can also be set—the large value is that the visual content belongs to the same-shot as the center of the interval, and the small value is that the visual content belongs to different shots. Or the weighting can also be set to - the large value is the visual atomic drop point near the center of the facet. The small value is the visual atomic drop point close to the facet boundary. The method of visual atom: as mentioned above, the visual atom describes the visual content of the temporary visual content. Data. In-execution, representing the time interval of the data, when the set of visual atoms is composed of the following steps: • Detecting a set of invariant feature points in the time interval; ' • Calculating the four ribs in each of the financial lions The set of the turning point area of the sewing material; • Eliminate the invariant feature points of a subset and its description; • Create a visual atom as the function and description of the remaining invariant feature points The method of feature detection: in addition to the feature detection method described above, the set of invariant feature points in the time interval in the visual data, the above calculation may use the corner detector or the benefit 49 4951832 Corner side n or fine day _ _ or use mser to perform the same method. If you use the MSER algorithm, it can be applied to a separate subset of video data or a subset of space and time that can be applied to video data. The description of the feature points may also be a SIFT description, a space-time coffee description or a § description. Tracking method: In the case of some real-time t, the calculation of the subset of the above description includes: the trajectory of the relatively invariant feature points in the time interval, The methods used are as follows: • Calculate a single-descriptor as the descriptor of the ship's point of the threat track; • Assign all descriptors of the feature points belonging to the track. This calculation function may be a descriptor of the invariant feature point. The median of the descriptors of the average or invariant feature points. ” Method of modifying feature features: In some embodiments, eliminating subsets as described above includes: • In time off The trajectory of the data for the invariant feature points; • Specify the quality matrix for each trajectory, • • = the invariant feature points on the trajectory whose quality matrix values are lower than the predetermined physiology.一些 一些 In some embodiments, the product functions assigned to the trajectory as described above are as follows: • Describe the value of the invariant feature point on the trace; • The drop point of the invariant feature point on the trace; proportional to shouting _ The difference between the forest and the forest (4) is the difference in the visual atomic structure: in the case of the symmetry of the six sounds, the creation of a visual atom as described above, the second atom, 50 200951832 :! The descriptor of the invariant feature point is taken as input; the cigarette smoke-doing fine and the best recognition feature point descriptor is received as a round-in; the index descriptor r of the representative descriptor found by the output can be used to The representative visual character of the canister _ method · 代 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ In some cases, it will be utilized to create a large array of views that are commonly used in all video's or at least in a large; 4' μ to promote the source of fresh resources. The method of seeing the atomic trimming is shown in some embodiments. In some embodiments, a subset of visual atoms, such as the set of visual atoms described above, may be associated with the age, and the subset of visual atoms that may be included: ❹ • in the set for each - A visual atom specifies a quality matrix; • A visual atom that eliminates a quality matrix value below a predetermined threshold. A threshold can be fixed or adjusted to maintain the minimum of the visual atom in the set. Or to minimize the maximum value of the visual atomic limit in the set. In addition, the specified quality matrix can include: • receiving a visual atom as input; • computing visual atoms and vector similarities representing a set of visual atoms; • output quality matrix as a function of similar vectors. This function may be proportional to the ratio of the maximum of the similar vector and the maximum of the similar vector to the second largest 51 200951832 value of the similar vector, or this function is the maximum of the similar vector and the maximum of the similar vector and the similarity of the orientation ^ The ratio of large values. Method of sequence calculus: In some embodiments, sequence calculations of visual nucleotides as described above may include: • receiving two visual nucleotide sequences 丨 and taking turns; • receiving a scoring equation σ(υ; And a gap penalty function, » as a parameter; • Find the corresponding part of the C = {(W丄 and gap function G function maximum value © *=1 • Output found part correspondence (: and function maximum) Other methods of counting the nose: As discussed earlier, maximization can be done using the Smith-Waterman algorithm, the Needleman-Wunsch algorithm, the BLAST, the '寅 algorithm, or possibly in a hierarchical approach. The scoring method: scoring The equation may be - or a plurality of functions as described above ^ Aqj ; s ·, ❹ λ / ^ · ^si j ^ (Jj where "may be a strange matrix, a diagonal matrix. , scoring may be proportional to (4) acid ^ The mutation is 4 nuclear thieves. The mutation probability may be estimated from the training experience of the training materials.” The score may also be proportional to the probability, and the dog change probability may be an empirical estimate of the training data. 52 200951832 From the basic scoring method: In addition, the 'scoring side light may be inversely proportional to the distance function 4?,,%) 'The possible combinations of the distance function and the distance function are at least as follows: • L1 distance; • Mahalanobis distance; • Kullback-Leibler divergence; • Earth Mover distance. Weighted calculation: In addition to the previously described weighting plan, the elements on the diagonal of the matrix can be proportional to "log", which means how many times the visual atom appears in the visual nuclear bitter acid. The video data of the training is estimated based on the input video data. The elements on the matrix read angle can also be proportional to the difference between 21 Κ -中 'V, which is the mutated version of the visual ζ ζ in the same visual (four) acid, ❹ ^ For the visual atom, there is a difference in any visual nuclear collision. In addition, the coffee may be estimated from the training video data. , The method of the two gaps: In some embodiments, the null separation can be a parameter of the formation function. In the beginning of the two sequences of .θ , , , A, and *7疋, “the gap length is a parameter. The parameters can be estimated from the training data, and the training data. Can encapsulate and delete the content. In addition, the gap penalty may form Γ7 ' &quot; represents the gap length and α and is the parameter. Step-by-step, to the gate = 疋 convex function ' or with the position at the beginning, and y The length of the gap is inversely proportional to the two sequences. 53 200951832 The method of spatial correspondence · The method of calculating the space _ should include: • input the time corresponding to the subset of video data; • provide feature points in a subset of the video material • Find the correspondence between feature points; • Find the correspondence between space coordinates. The temporal relative subset of the video material may be at least one pair of times. In addition, finding the correspondence between feature points is more likely to include: • inputting two sets of feature points; • providing descriptors of feature points; • matching descriptors. The feature points can be calculated by the 视 视 视 贼 ,, and the description is used for the same video nucleotide calculation. In addition, the other 'find the corresponding point between the feature points can use the algorithm or the parameter to describe the two _ mutual transformation feeding points, in the model of its number, · can solve the problem under the tactics ^ And the team a) is a set of two feature points, and τ is a parameter that converts two sets of feature points according to parameter 0. The correspondence between the '' and the two coordinates can be expressed as a mapping between the coordinates of the spatial system in the subset of video data &amp; (X ' y ) and the coordinates of the spatial system in another subset of video data (X, The method of 'the output of the time-space video data subsets is the coordinate (X, y, t) representing the 1 concentrated space-time system and the coordinates U, y, t of the other-subset space-time system. The mapping between '). , , 200951832 An example of the dna generation process for this video is shown in Figure 15. Here, a local feature detector is applied to image frames 1500 of various video sequences in a frame-wise manner. This feature detector finds a point of interest 15〇2, also referred to as a “feature point” in the video 'sequence. As previously discussed, many different types of 4-featured debt detectors can be used to include Harris corner detectors (c. Harris and M.

Stephens "A combined comer and edge detector", Alvey Vision Conference, 1988) 'Kanade-Lucas algorithm (BD Lucas and T. Kanade, 「 "An iterative image registration technique with an application to stereo vision", 1981) SIFT scale space Feature detector (DG LoWe,

Distinctive image features from scale-invariant keypoints, IJCV, 2004), etc. In general, the purpose of this feature detection algorithm is to design spatial distortions (eg, changing resolution, compressing noise, etc.) that characterize the feature sequence as robust or constant in the video sequence. To reduce transient noise and focus of the most useful feature points, the feature points typically leave a track 15〇4 in multiple faces, while feature points 1506 that occur too short will be deleted or trimmed. DNA The DNA generation process of the next stage video is displayed in the detail of a video image displayed in “16th image/here”, “16th image”. The image of the feature point at the point 15〇2 in the picture will be . Feature points 1600 after function trimming are used to describe a local feature descriptor. This feature descriptor generates a second type of vector representing the local neighboring region 16 of the feature point 1600 in the video plane. As is the case, many different algorithms can be used to describe the performance of video images near feature points. These algorithms include local histograms in the edge direction, = example invariant (10) sign conversion (SIFT), and accelerated point robustness (SURF) algorithms (H. Bay, τ·Tuytelaars and LG〇〇1, “plus ^ 叩R〇b(10) 55 200951832 features", 2006). /I In theory, the feature descriptor can represent a local region of the video image of the second type of vector ‘spot descriptor and the feature point side. His local neighborhood should be, the value of the second type of vector can be closer to trimming t!nr〇^ 1604 Yan two near t=?6=r two other J relative to the brightness of the corresponding image or tearing the stick, therefore, _ DNA thief Or describe the 6 denier of the visual ;; the identification of the ^ is not a short time sequence of the (8) 1 picture) contains two types of vectors of the 1. type - telling how many kinds of non-type = descriptives in the fragment t, (d) : Fine-to-marriage bribery, change the feature point to draw 2 hits - the difficulty of the standardization, can make many sly videos easier than two = unique descriptors with video segmentation, usually created in many not _video' The most appropriate description of "mapping" and "capacity two:" comes from any particular video entering this standardized data point 1600 (ϋ?" towel, as discussed earlier, for each trimming feature. The actual situation of the four-touch environment (4) gambling descriptor 1700 ^ material library or "visual word discarding is designated as "container", which is the type of pre-descriptor set. This visual discard can be regarded as a == = Fresh database. Here, I want to collect the feature m to represent the calculation of the feature details. An "ideal" feature descriptor in the visual database, which is closest to the real "feature descriptor. Therefore, each "real" feature description 56 200951832 is carried out from the actual face part ( Or instead of being in a visually close partner, and only the index closest to the "ideal" (ie, the _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ From the point of view of naming, the characteristic points of this method represent, sometimes, the "seeing atom" is regulated here. As a rough metaphor, the visual vocabulary can be seen ❹ 个The "weekly record" of atoms or molecules. Looking at the ship's metaphor, the visual vocabulary can be seen as a "periodic table" of visual elements. "18th picture" provides additional details showing how the original video is (time division) At this stage, the video sequence is segmented into different time (time) intervals or small segments, etc. _, delete, brain. Interval time can be a fixed size (such as 'every ten faces represent a range", or variable Size, also available It is overlapped or non-overlapping. It is often possible to chase _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Said, based on the feature point tracking from the previous period. It should mean that (4) the score is completed by the action of the pre-mosquito. Secondly, the visual vocabulary is now put into the visual feature descriptor (visual atom) to be combined in each intervening interval ( Aggregate) is the set 18〇6. Here, the time and space coordinates of the feature points themselves are not used, but in a series of video faces (time intervals), the sum of different types of feature descriptors is used. This process ends with the creation of a histogram, vector or signature (descriptor) for each series of video images. The various frequencies placed into the feature descriptor (visual atom) can be represented as a histogram or vector and used here. This histogram 57 200951832 or vector is sometimes referred to as a visual nucleotide. © The "Feature Pack" method has many advantages in extracting or indexing video. One of the advantages is that the _ method is stable and can be used to fine-tune the relationship between the videos, even if one or two videos are changed over the original face by overlapping pixels, the space clip ('for example, cropping) is changed to a different parsing Degree or picture rate, etc. For example, if ^ - a visual solution is modified (4), for example, in the original image, the new video sequence will include feature-mixing (a collection belonging to a heterogeneous video and other collections belonging to the overlay if the coverage is not very Large (ie most of the information on the screen is the original video) 'It still correctly matches the two from the two videos by the easy-to-match matching, determining the nuclear Wei (straight (four) or vector ^ points Matching is less than 100% correspondence between the two. "Figure 19" shows an example of a specific medium for Nacheng video DNA. Here, 'video DNA includes different time divisions (small fragments) taken from the video. Delete the sorted array, the "sequence" of the different "histograms", the "1" of the feature description j, or the "nucleus_", etc. The video can also be the original reference video of the metadata library on the device. Or the client video copied from the original reference video, the video can be extracted and indexed via the video DNA program. Generally, the video DNA created by the reference video can be similar to the video created by the client. So that the video can be used as an index or a frequency of DNA. XI, he sees this - reference video DNA creates an index, the established index allows other users to set up a reference video like a client The end copy, to find out the client, put ^^ reference video or in the health video library. For example, the client that plays the client video 1914 can calculate 1916 according to the same video 58 200951832 DNA program. The video DNA of the client video, the nature of the sequence of the client-side video Z's DNA identification device or other saved reference video dirty and the location can be based on the client video &lt;', s $ or &gt; The index of the video DNA database is determined. The index can then retrieve the data from the server, and the metadata corresponds to the portion of the video being played. As discussed earlier, the relative characteristics of the different features are Large arrays (ie ❹ = or thousands) can be analyzed by the image of the lining, and all the image feature points are placed in each __ feature point algorithm. Some image feature descriptors are not suitable for the input. The feature descriptor algorithm, or ambiguous adaptation. To correct the accuracy of the overall video DNA program, it is often useful to try to use the nearest neighbor algorithm to try to get the closest possible adaptation. In the nearest neighbor In the calculation ^, the actual observed feature points (feature descriptors) are created for use with the feature descriptor algorithm Qian Zhi and the observation ship close to the number of containers. The time matching and reference video DNA of the large customer ^ can be used. For a variety of different deductive algorithms. These algorithms range from very simple "match/mismatch algorithms" to biometric "like" algorithms, which are very complex algorithms. Biological DNA sequence matching for bioinformatics. Examples of these complex bioinformatics algorithms include the Needleman-Wunsch algorithm proposed by s. B. Needleman and CD Wunsch in 1970, by τ·F. Smith and MS.

The Smith-Waterman algorithm proposed by Waterman in 1981, the "identification of the sequence of general molecules" in 1981, and the "Basic Position Correction Search Tool (BLAST)" proposed by s F Alschul et al. in 199. Usually 'a suitable sequence matching algorithm will define a match score (or 59 200951832 distance)' to represent the quality of the match between the two video sequences. The matching score consists of two parts, the main part: the similarity (or distance) between the kernel _ and the vacant points, and the algorithm to comment on how to introduce gaps to avoid the "tearing" sequence. To do this, the distance between a nucleotide in the first video and the corresponding nuclear training in the second video is determined by the recording process. So, how is the "feature package" of the first server from the visual side and the "feature package" of the second feeding device from the second visual side? This similarity can be expressed as the measurement of two cores in a matrix. In the case of multiple similarities or multiple dissimilarities, it can represent the _ gamma distance or correlation of each material _ vector (abbreviated). If partial similarity is to be allowed (this often happens, especially if the visual nucleotic acid space is modified to contain non-feature points), a more complex weighting 2 or exclusion of non-subject data can be used. More complex distances may also be considered: i:, the probability of mutation of nuclear picric acid: two different nucleotide acids are very similar to each other. For example, consider the first view with a view. The frequency ^=4 two videos have the same video image as the first sequence of video tuner or pixels) will be similar to many of the points used in the feature package. In addition, the ulterior motive of the two feature points of the 鑤 肌 肌 导致 导致 为 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致 导致

For complex gap penalties, the number of skins may be multiplied by the number I. More .y will take into account the probability of a gap, for example, __ in the statistical distribution of the location of the port A. · According to i, although the "video job"-lion description method is a good method outlined in 200951832, it is obvious that matching various video nucleotides should be a simpler than matching the nucleotides of the biological nucleic acid complex. A", "T", "G, 戎" (- 1, J - human, and the DNA nucleotide of the video is a more complex "feature package" (the 'description of the feature descriptor'). So 'often one A particular video nucleotide will never find a phase roll • A perfect match. Instead, the standard "match" will usually be a very close, but not perfect match. Usually the 'match will be determined by the distance function. Such as distance, L1 distance, Mahalanobis distance, Kullback-Leibler divergence distance, Earth moving turbulence distance or other functions. This is an example, whenever the video nucleotide matches the "distance" &lt;= threshold value. Matching rules are considered to be a more rigorous match (ie a small number of video DNA nucleotides or identification tags will match each other), while larger matching rules are considered a less stringent match (That is, the DNA nucleotides or identification marks of more videos will match each other.) With regard to "20th to 24th", a series of illustrations illustrate a process configuration according to the system and the method described. Figure 2 shows an example of a video signature feature. In this example, the input video (A) is used to input a multi-ratio feature detector 2006 from a sequence and basis of video images containing video images 2004. The X and γ defined regions are formed during the time period. The video signals $, , .s2, and s3 are the rotations of the filters of different spatial depths (B), resulting in a series of different resolution characteristics. The ratio of images. These different ratios of spatial images will fall in different ratios 1, 2, 3 (C) after analysis (for example, corner detection). Pictures can be confirmed by the multi-ratio vertices (D) of the series. The feature points fl and β in the picture. The 21st picture of the factory represents an example of the tracking and trimming process of the feature points of the video symbol. This 200951832 is an optional stage, but if used, the feature points can be spanned across multiple pictures. The picture is sufficient (for example, the preset criteria are met) - the insured = the feature of the abandonment of the ridge is _ the number of coffees that do not have enough to meet the preset criteria, and the "22nd picture" shows an example of the video symbol feature descriptor. =_First side_How to describe after the feature point... In the normal case ^ - the person enters the video 22GG program, and this time analyzes the neighboring area around each of the previous ^^ feature points (6) in the video (x, y, r This feature = 2 = 7 (10) complement (four), - the week of the retweet _ the SIFT gradient of the near image is calculated (H), according to this gradient, the fineness of the ΪΓ = 7 twist direction (1) ° after this fine A vector (J) parsed into an element is called a feature descriptor. "23rd picture" shows an example of a process of vector quantization, which inputs a mapped image into a sequence of quantized feature descriptors. In this example, the video image is first ❹ = a feature descriptor vector (8) with an arbitrary feature descriptor vocabulary, which is mapped to a standardized d-dimensional interfering descriptor (1). —1〇Standardization Plan (SCh—(10)' The only video that confirms the source of the drug is not enough. The “24th picture” shows the creative analysis of video DNA, and the picture of the video is _ Feature package. An example of the Ting aspect is shown in Figure 8. ==? Video! and the feature package for the special face is a summary ": Ν) ι some feature packages from the video face (for example, Painting 62 200951832 face 2, picture 3) is averaged (P) to produce a representative of the multi-face video time interval, often referred to as "video nucleotides". Figure 25 shows the system for processing video data. Video source Save and/or generate video material. The video splitter 25〇4 receives the video data from the video data source 25〇2 and divides it into video data for the time interval. The video processor 25〇6 receives the video material from the video material source 2502 and performs various operations on the video material. In this example, video processor 25〇6 detects feature locations in the video material, generates feature inserts associated with the feature locations, and trims the detected feature locations to generate a subset of feature locations. Video aggregator 251G is coupled to video splitter 2504 and video processor 2506. The video aggregator 251 generates a video DNA about the video material. As discussed herein, the video DNA can include video material that is an ordered sequence of visual nucleotides. The storage device 25G8 is connected to the image splitter 25G4, and the video processor genus and the video concentrator 2MG can store a plurality of different materials used by the components. The stored data includes: video data, picture dragon, Guanlong, feature descriptor, visual ❿ Atomic, video DNA, algorithms, settings, thresholds, etc. The components described in "Dinitu" can be connected directly or via another intermediate device, system, component, network, communication link, etc. The system and method described by New is easy to confirm and correlate multiple video content identifiers associated with a particular video content. In addition, certain embodiments can be used with more than one conventional video processing and/or video display system and method. For example, an embodiment can be used to improve existing video processing systems. Although Nayong's metaphase and modularity are combined, the combination of these components and modules can be adjusted and associated with multiple video content identifiers in different ways. In other embodiments, one or more additional elements or modules may be added to the previously described system. Alternative embodiments may also combine two or more of the above-described elements or modules into one element or module. The invention disclosed in the above is the same as the above, but the content is not intended to be directly limited to the scope of the invention. Any modification of the form and details of the practice of the present invention may be made by those skilled in the art without departing from the spirit and scope of the invention. Special care for the fine. The special yarn of the present invention, the side of the collection, is subject to the definition of patent scope. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a schematic view of the environment proposed by the present invention. Figure 2 is a schematic diagram of a program executable by the universal video search system of the present invention. Figure 3 is a flow chart showing the extraction of the video content identifier and the meta-data related to the specific video content. Figure 4 is a flow chart for judging the correspondence between multiple video sequences according to the present invention. Figure 5 is a flow chart for confirming, capturing, and calibrating subtitle information according to the present invention. &gt; Figure 6A is a schematic diagram showing the spatial calibration and time calibration of the video material proposed by the present invention. Figure 6B is a schematic illustration of the context in which the video gene algorithm represents the context of the present invention. Figure 7 is a schematic diagram of the video DNA of the present invention. 64 200951832 Fig. 8 is a schematic diagram showing the comparison of DNA and video DNA of the organism of the present invention. Figure 9 is a flow chart of establishing a video DNA according to the present invention. Figure 10 is a schematic diagram of the present invention for dividing a video sequence into time intervals. Figure 11 is a flow chart showing the detection of feature points based on the face of the present invention. Φ Fig. 12 is a flow chart of the feature points for searching unchanged according to the present invention. Figure 13 is a flow chart of the trimming feature point trajectory proposed by the present invention. Figure 14 is a flow chart showing the temporal and spatial correspondence between the two video DNA sequences proposed by the present invention. Figure 15 is a schematic diagram of the video DNA generation program of the present invention. Figure 16 is a schematic illustration of how video feature points during trial frequency DNA generation are processed in accordance with the present invention. Figure 17 is a diagram showing how the video feature description symbols of the present invention are placed in the storage of the feature description symbols. Figure 18 is a schematic diagram of how the video of the present invention is 'segmented into a large number of short cells during the process of creating video DNA. • Figure 19 is a reference to the video DNA index and description related to the invention.

A schematic of the video. W Fig. 20 is a diagram showing the process of detecting a video recognition feature point according to the present invention. ° m ° ~ 21 is a sound diagram of the tracking and trimming video recognition feature points of the present invention. μ忍 65 200951832 Schematic. Figure 22 is a diagram showing the video identification feature point of the present invention. Figure 23 is a schematic diagram of vector quantization according to the present invention. Figure 24 is a construction video dna of the present invention

[Main component symbol description] Schematic. 102 104 106 108 110 112 114 116 118 202 204 206 208 210 212 214 216 600 First video source second video source third video source network _ 霸 first source data twentieth data source ternary data source universal Video search system video device communication module processor video analysis module © storage device video DNA module similar module · hash computing module advertising management module first stage second stage 66 602 200951832

610 Apple 612 Fruit 614 Computer 702 First Stage 704 Second Stage 706 Time Interval 708 Visual Element (Atomic) 710 Vision Range 712 Visual Nucleotide 714 Video DNA 800 Video Signal 802 Nucleotide and Atomic 900 Video Data 901 Video Data 1000 Feature Detection 1010 Feature Location 1020 Time Interval 1022 Time Interval 1024 Time Interval 1026 Time Interval 1425 Time Correspondence 1435 Subset 1445 Space Correspondence 1500 Image Face 67 200951832 1502 Feature Point 1504 Execution 1506 Deleted Meaningless Pendant 1600 Pruned feature points 1602 Local neighbors 1604 Feature descriptor 1700 Actual feature descriptor 1702 Representative feature descriptor i 1800 Time interval 1802 Time interval 1804 Time interval 1806 Feature set 1808 Time and space coordinates 1810 Feature package 1914 Client Video 1916 Client Video DNA Generation Program 2000 Feature Description 2004 Video Image 2006 Multi-Ratio Feature Detector 2010 Feature Descriptor 2200 Video 2500 Subsystem 2502 Video Data Source 2504 Video Splitter 68 200951832 2506 View Frequency Processor 2508 Storage 2510 Video Aggregator 3000 Feature Trimming * 3010 Feature Subset 3100 Feature Tracking 3110 Track 3200 Excavation Trim® 4000 Feature Representation 5000 Time Interval Segmentation 6000 Visual Atomic Aggregation 6010 Video DNA 6011 Video DNA 302 Receive and Special The video content related request 304 identifies the video content identifier associated with the specific video content 〇 306 retrieves the metadata associated with the particular video content based on the video content identifier - 308 310 312 402 404 406 Converting the video associated with the particular video content The content identifier is another video content identifier corresponding to the first video content identifier, and the metadata providing metadata related to each other video content identifier and the information corresponding to the visual reconciliation to the requesting device identification and video program The first video sequence identifies a second video sequence associated with the same video program to calculate a correspondence between the video sequence and the second video sequence. 69 200951832 408 4l 〇 412 502 504 506 508 510 512 514 determining the first video sequence and Calibration line storage between the second video sequences Storing the information about the calibration line to receive the specific subtitle information related to the specific movie in the DVD disc. The first DVD identifier associated with the DVD disc identifies the video sequence of the movie stored in the DVD disc according to the video sequence.顺 _ DVD _ phase H DVD $ 别 Use the first DVD identifier to listen to the subtitle source from the subtitle source to identify the correspondence between the DVD electric tree post and the movie timeline of the subtitle

According to two time, the school car _ subtitle information corresponds to the DVD

Claims (1)

200951832 VII. Patent application scope: 1. A method, the method comprising at least the following steps: receiving a request corresponding to a specific video content from a requesting device; confirming, by a video content source, one of the first corresponding to the specific video content a video content identifier; capturing, according to the first video content identifier, a first meta-data corresponding to the specific video content; &quot; converting the first-view_receiving job to an H-video content identifier; The second video content identifier captures one of the second meta-data corresponding to the specific video content; and “provides the first meta-data and the second meta-data to the requesting device. As described in claim II. The method of claim 1, wherein the first video content identifier corresponds to all of the specific video content. 3. The method of claim 1, wherein the first video content identification system and the method A specific time interval corresponding to one of the specific video contents. 4. The method of claim 100, wherein the first video content identifies 71 1 The method of claim 1, wherein the first video content identifier is an identifier based on the file name, and the identifier is stored in the identifier. The method of claim 1, wherein the first video content identifier is one of the identifiers based on the content of a video file. The identifier corresponds to the file bearing size of the specific video file. The method of claim 6, wherein the first video content identifier is based on the file bearing size of the specific video file. A method of calculating a hash value. 8. The method of claim 1, wherein the first video content identifier is an identifier based on the video content, the identifier and Obtaining information from the frame in the specific video content. 9. The method of claim 1, wherein the specific video content is The source is a Digitai versatile Disc (DVD), and the first video content identifier is a 〇^ identification code. The method of claim 1, wherein the specific The video content is obtained from a file of a peer-to-peer network, and the first video content identifier is a hash value corresponding to the file. The specific video content is obtained from a video file stored in a server, and the first video content identifier is one of identifying the server and the specific video file. The method of claim, wherein converting the first video identifier to a second video identifier comprises generating a second video identifier corresponding to all of the finalized video content. 13. The method of claim i, wherein the converting the first video., the sign-of-second step comprises: generating a _specific view _ · a time interval corresponding to the A second video identifier. The method of claim 13, wherein the converting the first, the step of distinguishing the second video identifier further comprises finding a time coordinate of a specific video material corresponding to the first video inner valley identifier The second video content 72 200951832 is a fine relationship. The method of claim 1, wherein the step of converting the first video to the second suffix comprises generating and framing the video, 16 such as the time space object - the second video identifier . The method of claim 15, wherein the step of converting the first video to: ,, and the second video of the video further comprises finding a space-time coordinate of the special frequency data corresponding to the first video in the video. And the second video content=extracting the relationship between the spatiotemporal coordinates of the specific video material. H Please refer to the method described in item 1, wherein the first meta-data is non-time metadata corresponding to the specific video content of the video. The method of claim 1, wherein the meta-metadata is associated with a time interval metadata of the _ interval in the special video content, as described in item i of the patent application scope. The method, wherein the first meta-data is a spatio-temporal metadata corresponding to a spatio-temporal object in the specific video content. .2〇=请!利范! In the second item, the method 'the (four)-meta data system is from the source of the first element, and the second element of the 5th is derived from the source of the second source. A method of claim 21, wherein the step of converting the first identifier to the second video identifier comprises performing a conversion according to each of the requests. 22. The method of claim </ RTI> wherein the step of converting the first video 73 200951832 identifier to the second video identifier further comprises a second video content identifier previously generated. The method of claim 1, wherein the step of confirming the first video inclusion step further comprises the step of receiving the first visual axis capacity identifier by the imaging device. 24. The method of claim 1, wherein the method further comprises the step of confirming a correspondence between a time axis corresponding to the specific video content and a time axis corresponding to the second metadata. . 25. The method of claim 24, wherein the method further comprises the step of providing the correspondence to the requesting device. a method comprising at least one of: determining a first video sequence associated with a video program; identifying a second video sequence associated with the video program; calculating the first video sequence and the second video Corresponding relationship between the sequences; determining a calibration line of the first video sequence and the second video sequence; storing the correspondence between the first video sequence and the second video sequence; and storing the first video sequence And the information of the calibration line between the second video sequence. 27. The method of claim 26, wherein the step of calculating the first video sequence and the second video sequence comprises calculating the first video sequence and the second video sequence One of the spatial correspondences. 28. The method of claim 26, wherein the step of calculating the correspondence between the sequence of the video-video 74 200951832 and the second video sequence comprises calculating the first video sequence and the second video sequence One of the time correspondences. 29. The method of claim 26, wherein the calculating the correspondence between the first video sequence and the second video sequence comprises calculating the first video sequence and the second video sequence A time-space correspondence. 30. The method of claim 26, wherein the step of determining the calibration line of the first video sequence and the second video sequence comprises calculating video-based complex descriptors, The description symbol corresponds to the first video sequence and the second video sequence. 31. A method of at least the method comprising: receiving a request for one of a meta-data associated with a video program, the request including a first video content identifier corresponding to one of the video programs; a video sequence; confirming, according to the video sequence, a second video content identifier corresponding to one of the video programs; 撷 extracting the metadata from the unary data source based on the second video content identifier; and * by one of the corresponding video programs A time axis and a corresponding relationship with the time-axis of one of the corresponding metadata is confirmed. 32. The method of claim 31, wherein the step of confirming the correspondence comprises calculating a spatial correspondence between the video program and the metadata. 33. The method of claim 31, wherein the step of confirming the correspondence comprises calculating a time correspondence between the video program and the metadata. 0 75 200951832 34. The method, wherein the step of confirming the correspondence includes calculating a space-time correspondence between the video program and the metadata. 35. The method of claim 31, wherein the step of confirming the correspondence comprises the step of extracting a previously calculated correspondence from a storage device. The method of claim 31, wherein the step of confirming the correspondence comprises calculating a calibration line between the video program and the metadata. 37. The method of claim 31, wherein the meta-data is subtitle information corresponding to the video program. 〇 38. The method of claim 31, wherein the meta-data further includes a comment on the video program. 39. A device, the device comprising: - a communication module 'for receiving a request for a specific video content from a requesting device; and - a processor 'connecting with the communication module' to confirm that the corresponding video is corresponding Content - the first - visual _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ The second-level information is obtained according to the second video content identifier of the Wei, and the information and the second-dimensional data are further provided to the requesting device. 4. The device according to claim 39, wherein the towel is used to find the time coordinate of the specific video material corresponding to the first video content and the second video content identifier. Corresponds to the relationship between time coordinates of a particular video material. 76. The apparatus of claim 39, wherein the processor is further configured to find the age coordinate of the miscellaneous fixed woven fabric and the second video content identification. The relationship between the space-time coordinates of a particular video material. 1 42. A method comprising the steps of: inputting a first video identifier, the first video identifier comprising a content identifier and a first set of coordinates in a first video content; a second video identifier of a second video content, wherein 'the second video content is similar to the first video content; and performing the following steps for each video identifier in the second set of video identifiers: confirming the a set of coordinates in the second video content; and outputting the set of coordinates in the second video content to request metadata of one of the first video content. 43. The method of claim 42, wherein the first coordinate of the group is a set of time points in a time axis, wherein the time axis corresponds to the first video content. The method of claim 42, wherein the first coordinate of the group is a set of time intervals in a time axis of the first video content. 45. The method of claim 42, wherein the first coordinate of the group is a set of spatial coordinates&apos; wherein the spatial coordinate corresponds to one of the first video content. 46. The method of claim 45, wherein the set of spatial coordinates is a diagonal comers of a bounding box. 77. The method of claim 42, wherein the first coordinate of the group is a set of space-time coordinates, wherein the space-time coordinate corresponds to one of the first video content. 48. The method of claim 47, wherein the set of space-time coordinates is a set of time intervals, and space coordinates corresponding to the objects are respectively provided to each of the time intervals. 49. The method of claim 42, wherein the step of confirming the second set of video identifiers comprises extracting the second video identifier from a table, wherein the table includes one of storing video identifiers The first block, and the second field of the video identifier associated with the video content that is similar to the video content associated with the video identifier in the first block. 5. The method of claim 49, wherein the form is pre-computed. 51. The method of claim 49, wherein the form is a correspondence table. 52. The method of claim 49, wherein the step of calculating the form comprises calculating a description symbol based on the video content in the first video content, and comparing the description symbol with The content base description in a database. The symbol identifies other video content similar to the first video content. 53. The method of claim 52, wherein the content base description symbol is a visual nucleotide (visual nueleQtide). The device comprises at least: ^ a communication module, For receiving a first video identifier, the first video identifier includes a content identifier and a set of first 78 200951832 coordinates in a first video content and used to transmit a second video identifier and The second video identifier corresponds to a second set of coordinates, the second coordinate of the set is in a second video content, and a processor is connected to the communication module for extracting the first The second video axis character 'has the second vision pain of the lining, the silk screen includes a first block for storing the video identifier, and the video content storing the video content related to the video identifier in the first block The second field of the associated video identifier. ❹ 79