CN110543584B - Method, device, processing server and storage medium for establishing face index - Google Patents

Abstract

The embodiment of the invention provides a method, a device, a processing server and a storage medium for establishing a face index, wherein the method comprises the following steps: acquiring at least one video; respectively determining corresponding face features of each video, and determining video information of each face feature in the corresponding video; clustering the face features of the same face to obtain at least one first cluster, wherein the face features aggregated by the first cluster represent the face features of the same face in the at least one video; and aiming at each face feature gathered by each first cluster, correlating video information of the face feature in the corresponding video to obtain the face index of at least one video. The embodiment of the invention can efficiently establish the face index under the condition of no face information mark.

Description

A method, device, processing server and storage medium for establishing a face index

technical field

The invention relates to the technical field of data processing, in particular to a method, device, processing server and storage medium for establishing a face index.

Background technique

The face index represents the association between the face features in the video and the video information. By establishing a face index for the video, it can efficiently provide the video information associated with the target person in the video when querying scenes such as the target person in the video. (such as the video time point when the target person appears in the video, the video progress, etc.); based on the characteristics of the face index, the face index has been widely used in video on demand, security and other fields.

At present, the establishment process of the face index generally requires the user to mark the face information in advance; however, there are often many people in the video, and the establishment of the face index based on the way the user marks the face information will be extremely cumbersome. The work of information, and the difficulty of marking face information is also extremely difficult, which undoubtedly leads to the extremely low efficiency of face index establishment.

Contents of the invention

In view of this, the embodiments of the present invention provide a method, device, processing server, and storage medium for establishing a face index, so as to efficiently establish a face index when there is no face information tag.

In order to achieve the above purpose, embodiments of the present invention provide the following technical solutions:

A method for establishing a face index, comprising:

Get at least one video;

Respectively determine the corresponding facial features of each video, and determine the video information of each facial feature in the corresponding video;

The facial features of the same face are clustered to obtain at least one first cluster; wherein, the facial features gathered by a first cluster represent the facial features of the same face in the at least one video;

For each face feature gathered by each first cluster, associate the face feature with the video information of the corresponding video, and obtain the face index of the at least one video.

The embodiment of the present invention also provides a device for establishing a face index, including:

A video acquisition module, configured to acquire at least one video;

Facial features and video information determining module, used to respectively determine the corresponding facial features of each video, and determine the video information of each facial feature in the corresponding video;

The first clustering obtaining module is used to cluster the facial features of the same face to obtain at least one first cluster; wherein, the facial features gathered by one first cluster indicate that the at least one video Facial features of the same face in the

The face index establishment module is used for associating the video information of the face feature in the corresponding video with respect to each face feature gathered by each first cluster, so as to obtain the face index of the at least one video.

An embodiment of the present invention also provides a processing server, including: at least one memory and at least one processing chip, the memory stores a program, and the processing chip calls the program to implement the above-mentioned method for establishing a face index A step of.

The embodiment of the present invention also provides a storage medium, the storage medium stores a program suitable for execution by the processing chip, so as to realize the steps of the above-mentioned method for establishing a face index.

Based on the above technical solution, the method for establishing a face index provided by the embodiment of the present invention can perform clustering processing of the face features of the same face on at least one video, and obtain the corresponding first clustering method suitable for the at least one video. Classes; and then each of the face features gathered by each first cluster is associated with the video information of the corresponding video, so that it can be efficiently established in the case of no face information mark. Face index of a video.

Further, by clustering the face features of the same face in the at least one video, it can be considered that the face features gathered in a first cluster represent the same face, which can reduce the number of faces in the face index library. The number of faces can greatly reduce the data redundancy of the face index.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only It is an embodiment of the present invention, and those skilled in the art can also obtain other drawings according to the provided drawings without creative work.

Fig. 1 is a traditional schematic diagram of establishing a face index based on user-marked face information;

FIG. 2 is a schematic structural diagram of a system for establishing a face index provided by an embodiment of the present invention;

FIG. 3 is a signaling flow chart of a method for establishing a face index provided by an embodiment of the present invention;

Fig. 4 is under the situation of a plurality of videos, establishes the processing example diagram of face index;

Fig. 5 is another signaling flowchart of the method for establishing a face index provided by an embodiment of the present invention;

Fig. 6 is an example diagram of processing of establishing a face index in the case of a video;

7 is a flow chart of a method for clustering facial features of the same face provided by an embodiment of the present invention;

FIG. 8 is a flow chart of another method for clustering facial features of the same face provided by an embodiment of the present invention;

Fig. 9 is a schematic diagram of adding face features to clustering;

Fig. 10 is a schematic diagram of clustering of facial features for deleting secondary faces;

Fig. 11 is an application example diagram provided by an embodiment of the present invention;

FIG. 12 is a structural block diagram of a device for establishing a face index provided by an embodiment of the present invention;

Fig. 13 is a structural block diagram of a processing server provided by an embodiment of the present invention.

Detailed ways

Figure 1 is a traditional schematic diagram of building a face index based on the user's marked face information. face image (such as a person's front face image, side face image, etc.), so that the server uses the marked face information as an input for registration, and through the registration process, determines the facial features of the face image in the video Corresponding video information, and associating the video information with the basic information and face features of the person, to establish the face index of the person;

It can be seen that when there are many people in the video (especially for large-scale video data) in the traditional way of building a face index, there will be extremely cumbersome work for users to mark face information, and the difficulty of marking face information is also extremely high. Large, there is a problem that the establishment efficiency of the face index is extremely low;

Furthermore, for people who are not marked with face information but exist in the video, the traditional way of building a face index is to use the face image without marked face information in the video as an independent input for registration processing, which makes people The data of the face index is extremely redundant.

In order to solve the above-mentioned defects, the embodiment of the present invention provides a scheme for establishing an efficient face index in the case of no face information tag; The technical solution is clearly and completely described, and obviously, the described embodiments are only some embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

FIG. 2 is a schematic diagram of an optional architecture of a system for establishing a face index provided by an embodiment of the present invention. As shown in FIG. 2 , the system may include: a video source 10, a processing server 20, and a face index library 30.

Wherein, the video source 10 can be regarded as the source of the video, depending on the application scenario of the embodiment of the present invention, optionally, the video can be represented by streaming data, such as a video stream;

As an example, for a live broadcast scene, the video source may be a live video server, and the embodiment of the present invention may establish a face index for the live video output by the live video server; for an on-demand scene, the video source may be an on-demand Video library (multiple videos are recorded in the on-demand library), the video-on-demand library can provide the video requested by the user when the user orders the video, and in the embodiment of the present invention, the video recorded in the video-on-demand library can also be processed manually. The establishment of face index; Of course, the video source can also be a movie video library, a TV video library, etc., which will not be explained one by one; obviously, the video source can be at least one (one or more) of the video source forms described above.

The processing server 20 is a service device for setting up a face index to the video provided by the video source 10, and is the main processing device for setting up a face index in the embodiment of the present invention; the processing server can be realized by a single server, or can be composed of multiple servers group implementation.

The face index library 30 is a database for recording the face index established by the embodiment of the present invention.

As a variant of the system structure, the face index database can also be implemented by a storage unit in the processing server, for example, a storage device with data storage capability in the processing server can record the face index established by the embodiment of the present invention.

In order to realize efficient establishment of a face index in the absence of face information marking, the core process of establishing a face index in the embodiment of the present invention may be as follows:

The processing server obtains at least one video for establishing a face index sent by the video source; for the at least one video, the processing server can determine the corresponding facial features of each video, and determine that each facial feature is in the corresponding video. Video information (such as each face feature at the video time point of the video to which it belongs, and/or video progress, etc.); thus, the processing server can cluster the face features of the same face to obtain at least one first cluster ( For ease of description, in this embodiment of the present invention, a cluster of facial features of the same face in the at least one video can be defined as a first cluster; that is, the facial features gathered by a first cluster can represent , the face feature of the same face in the at least one video); and then, for each face feature gathered by each first cluster, associate the face feature in the video information of the corresponding video, and obtain the video information suitable for the at least one video. Video face index.

Based on the above-mentioned core process, the processing server can perform clustering processing of facial features of the same face on at least one acquired video without any face information mark, to obtain at least one first cluster, so that each first Each facial feature of clustering gathers, carries out the correlation of facial feature in the video information of corresponding video, obtains the human face index of described at least one video; Realize under the situation of no face information mark, efficiently establish human face index .

Based on the above core process, the embodiment of the present invention can establish a face index for one video, and can also establish a face index for multiple videos. The two cases will be described below.

Optionally, Figure 3 shows an optional signaling process for establishing a face index provided by an embodiment of the present invention in the case of establishing a face index for multiple videos; it is worth noting that, as shown in Figure 3 The process is only an optional process for establishing a face index in the case of multiple videos in the embodiment of the present invention. Based on the core process described above, the embodiment of the present invention can also use other methods in the case of multiple videos. Process to establish face index;

Referring to Fig. 3, when the face index is established based on multiple videos, the process provided by the embodiment of the present invention may include:

Step S10, the video source inputs multiple videos to the processing server.

The multiple videos referred to here can be the multiple videos that the video source has, and can be considered as the multiple videos used to establish the face index in the embodiment of the present invention, that is, the embodiment of the present invention can be the integrated video of the multiple videos. Create a face index;

As an example, the video source can randomly input multiple videos to the processing server. For example, the video source randomly inputs multiple videos that the video source has to the processing server. Video, comprehensively builds a face index; optionally, the video source can also be specified (such as specified by the processing server, or specified by a staff member), and input multiple specified videos to the processing server;

Optionally, in the embodiment of the present invention, each video that the video source has may have a unique identifier (such as a video ID);

Optionally, any video input from the video source to the processing server can be in the form of a stream (i.e. video stream), and correspondingly, the video source can input multiple video streams to the processing server; obviously, the existence of video in the form of a stream is only possible Optionally, the embodiments of the present invention may also support face index building for other forms of videos. For example, after the processing server acquires multiple complete videos from the video source, it may perform face index building on the complete multiple videos.

Correspondingly, the processing server can obtain multiple videos.

Step S11, the processing server respectively determines the corresponding facial features of each video, and determines the video information of each facial feature in the corresponding video.

The processing server can respectively determine corresponding facial features for each piece of video obtained, and determine the video information of each determined facial feature in the corresponding video (such as the video time point of the corresponding video, and/or video progress, etc.);

As an example, for any video obtained by the processing server, the processing server can extract facial features from key video frames in the video, so as to process each video obtained by the processing server, and determine the corresponding person in each video. Face feature; Simultaneously, for the corresponding human face feature of each determined video, determine the corresponding key video frame of each human face feature in the video information of the corresponding video (as the corresponding key video frame of each human face feature, in the belonging video video timing and/or video progress);

For example, for a certain video acquired by the processing server, the processing server may extract the facial features of the key video frames in the video, and determine the video information of the key video frames of each facial feature in the video.

As another example, for any video obtained by the processing server, the processing server may take screenshots of the video according to a set time interval, and extract facial features from each screenshot, so as to process each video obtained by the processing server, Determine the corresponding face features of each video; at the same time, determine the corresponding screenshots of each face feature in the video information of the corresponding video (such as the corresponding screenshot of each face feature, the video time point and/or video progress in the corresponding video );

For example, for a certain video acquired by the processing server, the processing server may take screenshots of the video according to a set time interval, extract the facial features of each screenshot, and determine the video information of the screenshot of each facial feature in the video.

Optionally, the above two methods of determining the corresponding facial features of each video can be used in combination, or one can be used; of course, the embodiment of the present invention can also be used for each video frame of each video acquired Perform facial feature extraction to determine the corresponding facial features of each video; at the same time, determine the video information of the video frame corresponding to each facial feature in the corresponding video.

Optionally, the video information of the face feature in the corresponding video may also include: an identifier (such as a video ID) of the video corresponding to the face feature.

Step S12 , the processing server clusters the facial features of the same face with respect to the corresponding facial features of each video, so as to obtain at least one second cluster corresponding to each video.

After determining the corresponding facial features of each video, for any video, the embodiment of the present invention can cluster the facial features of the same human face in the video into one category to obtain the corresponding second cluster of the video (i.e. a A second clustering of video can represent, a clustering of the face feature of same face in this video); Each video is all processed with this, then can obtain the corresponding second clustering of each video;

It can be understood that a first cluster refers to a cluster of facial features of the same face in at least one video used by the processing server to create a face index; In the case of obtaining multiple videos, a cluster of facial features of the same face in one video.

For any video, the embodiment of the present invention can analyze the similarity between the facial features of the video, and group the facial features whose similarity meets the predetermined similarity requirements in the video into one category, so as to realize the same human face in the video Clustering processing of facial features;

Optionally, the predetermined similarity requirement can be determined according to the actual situation, and the embodiment of the present invention is not limited. As an optional implementation, the facial features can be represented by human facial feature vectors (such as using high-dimensional facial feature vector representations face features), the similarity between face features can be represented by the distance between face feature vectors, and the corresponding predetermined vector distance can be set to represent the face features of the same face, and the distance between face feature vectors can be set in the predetermined vector The face features within a distance are clustered into one class.

Further, there are often main human faces and secondary human faces in a video, and for any video, the embodiment of the present invention can determine the clustering of the facial features of each main human face in the video (the number of main human faces in a video may be one or more), to obtain the corresponding second cluster of the video;

Correspondingly, when the face features of the same face are clustered for the corresponding face features of each video, for any video, the embodiment of the present invention can cluster the faces of the same face in the video. After feature clustering, delete the cluster corresponding to the secondary face in the cluster of the video, and only keep the cluster corresponding to the main face in the cluster of the video to obtain the corresponding second cluster of the video;

As an optional implementation, for any video, the embodiment of the present invention can cluster the facial features of the same face in the video, and then further cluster the video with the number of occurrences of the facial features in the cluster less than the number threshold. Class is deleted (that is, the cluster of the secondary face in the cluster of the video is deleted), and the corresponding second cluster of the video is obtained.

Step S13 , the processing server correlates the facial features gathered by the second clusters corresponding to each video with the video information of the corresponding video, and obtains the clustering results of each video.

After determining the corresponding second clusters for each video and obtaining at least one second cluster corresponding to each video, the embodiment of the present invention can respectively focus on the people gathered by the second clusters corresponding to each video. Face feature, carries out the association of face feature in the video information of corresponding video; As for any second clustering (a second clustering represents, the clustering of the face feature of same face in a video), the present invention implements For example, each face feature gathered by the second cluster can be associated with the video time point and/or video progress of the video to which the face feature belongs; in this way, the corresponding second cluster of each video can be processed to obtain The clustering results of each video;

It can be seen that the clustering result of a video may at least include: at least one second cluster that gathers the facial features of the same face in the video, and each facial feature gathered by each second cluster in the video. video information.

It should be noted that step S12 and step S13 are only an optional method for obtaining the clustering result of each video in the embodiment of the present invention, and the embodiment of the present invention may also use other methods to obtain the clustering result of each video.

Step S14 , the processing server clusters the facial features of the same face for the clustering results of each video, and obtains at least one first cluster corresponding to the plurality of videos.

After obtaining the clustering results of each video, the second clustering of the facial features of the same face can be clearly defined for each video; The facial features of the human face are clustered, and the clustering of the facial features of the same human face suitable for the multiple videos is determined to obtain the corresponding first cluster of the multiple videos (that is, the at least one When the video is multiple videos, the clustering of the facial features of the same face in the multiple videos).

Step S15 , for each facial feature gathered by each first cluster, associate the facial feature with the video information of the corresponding video, and obtain the face index of the at least one video.

After obtaining the corresponding first clusters of the multiple videos, for each facial feature gathered by each first cluster, the facial features can be associated with the video information of the corresponding video to obtain the facial features of the multiple videos. face index.

Step S16, the processing server writes the face index into the face index library.

The method for establishing a face index provided by the embodiment of the present invention can perform clustering processing of the facial features of the same face for each video in the case of multiple videos, and obtain the corresponding second clustering of each video. Class, by the face feature that each video corresponding second cluster gathers, carries out the correlation of face feature in the video information of corresponding video, can obtain the clustering result of each video; Thereby each can be further In the clustering result of the video, the face features of the same face are clustered again to obtain the corresponding first clusters suitable for the plurality of videos; and then the face features gathered by each first cluster, By associating the face features with the video information of the corresponding video, an efficient establishment of a face index suitable for the at least one video can be realized in the absence of a face information tag.

For ease of understanding, multiple videos are divided into two videos and divided into the first video and the second video as an example. Assume that the first video and the second video have the same person A, and the first video has the same person A as the second video. Different characters B, there is a character C different from the first video in the second video (optional, characters A and B can be the characters of the main faces in the first video, and characters A and C can be the main people in the second video face), the corresponding processing example can be shown in Figure 4:

After the processing server acquires the first video, it can extract facial features from key video frames in the first video and/or screenshots with a set time interval, and perform clustering processing on the facial features of the same face to obtain the first The second cluster FA of the face features of person A in a video is represented by [FA1, FA2, ..., FAn], and each face feature of person A is at the corresponding time point TA of the first video, using [TA1, TA2,...,TAn] represent; obtain the second clustering FB of the face feature of character B in the first video, use [FB1, FB2,...,FBn] to represent, and each face feature of character B in the first video The corresponding time point TB is represented by [TB1, TB2, ..., TBn];

And, after the processing server acquires the second video, it can extract facial features from key video frames in the second video and/or screenshots with a set time interval, and perform clustering processing on the facial features of the same human face, Obtain the second cluster FA' of the face features of person A in the second video, represented by [FA1', FA2',..., FAn'], and each face feature of person A at the corresponding time point of the second video TA', represented by [TA1', TA2',..., TAn']; the second clustering FC of each face feature of character C in the second video is obtained, represented by [FC1, FC2,..., FCn], and Each face feature of character C is represented by [TC1, TC2, ..., TCn] at each time point TC corresponding to the second video;

After the processing server clusters the facial features of person A in the first video into one category, and the facial features of person B into one category, it can target the second clustering of the facial features of person A in the first video. Each face feature is associated with each time point corresponding to the first video, and the ID of the corresponding first video (the ID of the first video can be represented by VID1), and the clustering result of character A in the first video is obtained, which can be Use [<FA1, VID1, TA1>, <FA2, VID1, TA2>..., <FAn, VID1, TAn>] to represent; face features of the second cluster for the face features of person B in the first video , associated with each time point corresponding to the first video, and the corresponding ID of the first video, to obtain the clustering result of character B in the first video, you can use [<FB1, VID1, TB1>, <FB2, VID1, TB2>…, <FBn, VID1, TBn>] means;

And, after the processing server clusters the facial features of person A in the second video into one group, and after the facial features of character C are grouped into one group, the second clustering of the facial features of character A in the second video can be Each face feature of the class is associated with each time point corresponding to the second video, and the ID of the corresponding second video (the ID of the second video can be represented by VID2), and the clustering result of person A in the second video is obtained , can be represented by [<FA1', VID2, TA1'>, <FA2', VID2, TA2'>..., <FAn', VID2, TAn'>]; The face features of the two clusters are associated with the corresponding time points of the second video and the ID of the corresponding second video to obtain the clustering result of the character C in the second video, which can be used [<FC1, VID2, TC1>, <FC2, VID2, TC2>…, <FCn, VID2, TCn>] means;

After the processing server obtains the clustering results of character A and character B in the first video, and obtains the clustering results of character A and character C in the second video, the processing server can combine the first video and the second video The face features of the same face in the second video are clustered, that is, the face features in the clustering result of person A in the first video are clustered with the face features in the clustering result of person A in the second video Class processing, thereby obtaining the first clustering of the person A suitable for the first video and the second video, using [<FA1, FA2, ..., FAn>, <FA1', FA2', ..., FAn'>] to represent; Since characters B and C are different characters in the first video and the second video, the second clustering of character B in the first video can be used as the first clustering of character B suitable for the first video and the second video, The second cluster of person C in the second video may serve as the first cluster of person C applicable to the first video and the second video;

Furthermore, each face feature gathered by each first cluster can be associated with the video information of the corresponding video to obtain a face index suitable for the first video and the second video; that is, the obtained face index can be Expressed as: [<FA1, VID1, TA1>, <FA2, VID1, TA2>…, <FAn, VID1, TAn>, <FA1’, VID2, TA1’>, <FA2’, VID2, TA2’>…, <FAn', VID2, TAn'>], [<FB1, VID1, TB1>, <FB2, VID1, TB2>…, <FBn, VID1, TBn>], [<FC1, VID2, TC1>, <FC2, VID2, TC2>..., <FCn, VID2, TCn>].

Optionally, the second cluster FA of the face features of person A in the above-mentioned first video can also be represented by the mean value of [FA1, FA2, ..., FAn], and the face features of person B in the above-mentioned first video The second cluster FB can also be represented by the mean value of [FB1, FB2, ..., FBn], and the second cluster FA' of the facial features of person A in the second video can also be represented by [FA1', FA2' , ..., FAn'] means that the second cluster FC of the face features of person C in the second video above can also be represented by the mean of [FC1, FC2, ..., FCn].

The above shows the scheme of establishing a face index in the case of multiple videos; the embodiment of the present invention can also establish a face index in the case of a video; optionally, Figure 5 shows the embodiment of the present invention Another optional signaling process for establishing a face index is provided; referring to FIG. 5, when establishing a face index based on a video, the process provided by the embodiment of the present invention may include:

Step S20, the video source inputs video to the processing server.

The video referred to here can be any video that the video source has, and can be considered as any video used to establish a face index in the embodiment of the present invention;

As an example, the video source can input video to the processing server one by one or randomly, such as the video source can input the video that the video source has to the processing server one by one or randomly, then in this process, the processing server can input any video source. One video, build face index one by one; Optionally, the video source can also be specified (such as specified by the processing server, or specified by the staff), and input the specified video to the processing server;

Optionally, in the embodiment of the present invention, each video that the video source has may have a unique identifier (such as a video ID);

Optionally, the video may also exist in the form of a video stream.

Correspondingly, the processing server can obtain the video.

Step S21, the processing server determines the facial features corresponding to the video, and determines the video information corresponding to each facial feature in the video.

Optionally, for a video, the manner of determining the corresponding facial features of the video may refer to step S11 in FIG. 3 .

Step S22 , the processing server clusters the facial features of the same face for each facial feature corresponding to the video to obtain a first cluster corresponding to the video.

Optionally, for a video, the process of clustering the face features of the same face among the corresponding face features of the video can be shown in step S12 in FIG. 3 ;

It should be noted here that in the case of multiple videos, the clustering of the face features of the same face in each video is called the second cluster, and the same face of each second cluster in the multiple videos The face feature clustering can be referred to as the first clustering suitable for the multiple videos (a form of the first clustering referred to in the embodiment of the present invention in the case of multiple videos);

And in the case of a video, the clustering of the facial features of the same face in the video can be considered as the first cluster suitable for the video (the first cluster referred to in the embodiment of the present invention is in the case of a video) a form of).

Step S23 , the processing server correlates the face features of each cluster corresponding to the video with the video information of the video, and establishes a face index of the video.

After clustering the facial features of the same face in the video to obtain the corresponding clusters of the video, the embodiment of the present invention can associate the facial features of the corresponding clusters of the video with The face feature is in the video information of the video, and the corresponding clustering result of the video is obtained; thus, when a face index is established for a video, the corresponding clustering result of the video can be used as the clustering result of the video facial features.

It can be seen that, in the embodiment of the present invention, the corresponding clustering result of a video may include: each cluster of the facial features of the same face in the video, and the corresponding video information of each facial feature of each cluster in the video .

Optionally, the video information of a clustered face feature in the corresponding video may at least include: each face feature of the cluster at the time point of the corresponding video; as an alternative, a clustered face feature in The video information of the corresponding video may include: the video progress of each face feature of the cluster in the corresponding video; obviously, the video information of a cluster of face features in the corresponding video may also include: each face feature of the cluster at the point in time of the corresponding video, and/or the progress of the video;

Further, the video information may also include: a video identifier, such as a video ID.

As a hypothetical example, in the case of a video, if F represents the facial feature (such as using a high-dimensional facial feature vector), T represents the corresponding time point of each facial feature in the video, if the video There is a main face, then the first clustering of the facial features of the main face in the video can be obtained, expressed by [F11, F12, ...,, F1n]; at the same time, each of the first clustering can be determined The face features are at the corresponding time points [T11, T12, ..., T1n] of the video; for each face feature of the first cluster, after the associated face features are at the corresponding time points of the video, under a video, An alternative representation of a face index could be:

[<F11, F12, ...,, F1n>, <T11, T12, ..., T1n>]; where, <F11, F12, ...,, F1n> can also be represented by the mean value of face features; optional, the The face index may also include: the identification of the video;

Of course, when there are multiple faces (such as multiple main faces) in a video, in the above-mentioned face index, there may be multiple first clusters of face features belonging to different faces, and each first cluster Each face feature of the class is associated with the corresponding time point in the video.

Step S24, the processing server writes the face index into the face index database.

In the case of building a face index for a video, after obtaining the corresponding clustering result of the video, the embodiment of the present invention can use the corresponding clustering result of the video to realize the facial features of the same face in the video. Clustering, and associating the face features of each cluster in the corresponding video information of the video, when no face information is marked, the face features of at least one identical person in the video are associated with the video information to achieve The establishment of the face index of the video.

And by clustering the face features of the same face, it can be considered that the face features of the same cluster represent the same face, which can reduce the number of faces in the face index database, which can greatly reduce the number of faces. Data redundancy for face indexing.

For ease of understanding, in the case of a video, the corresponding processing process can be shown in Figure 6. The video information is represented by the face feature at the corresponding video point of the video. Referring to Figure 6, the person with the same face in the video is After the face features are clustered, a plurality of clusters (that is, multiple first clusters corresponding to the video) can be obtained; wherein, cluster 1 can be the face features of the same face <F11, F12, ..., F1n > clustering, and correlating each face feature in cluster 1 at the corresponding video time point of the video, [<F11, T11>, <F12, T12>..., <F1n, T1n>] can be obtained, where, <T 11, T12, ..., T 1n> means [F11, F12, ..., F1n] respectively corresponding time points in the video;

Cluster 2 can be the clustering of the face features [F21, F22, ..., F2n] of another face, and correlating each face feature in cluster 2 at the corresponding video time point of the video, [< F21, T21>, <F22, T22>..., <F2n, T2n>], where <T21, T22,..., T2n> represent [F21, F22,..., F2n] respectively corresponding time points in the video.

The above describes the establishment of face index based on multiple videos, and the establishment of face index in the case of one video; generally speaking, the embodiment of the present invention aims at at least one video for which a face index needs to be established, and the same face The facial features of each first cluster are clustered to obtain the first cluster suitable for the at least one video, and the facial features of the first clusters are associated with the video information of the corresponding video to establish a cluster suitable for the at least one video. The face index; Refer to note that said at least one video can be a plurality of videos (the processing situation shown in Figure 3 above), or a video (the processing situation shown in Figure 5 above);

Of course, in addition to the above-mentioned processing situations shown in FIG. 3 and FIG. 5 , in the case of multiple videos, the embodiment of the present invention can also cluster the facial features of the same face in the multiple videos without For each video, the facial features of the same face in each video are clustered.

Optionally, in the implementation process of clustering the same facial features among the facial features of the video, the embodiment of the present invention can set a predetermined similarity requirement, and by analyzing the similarity between facial features, similar The face features whose degrees meet the predetermined similarity requirements are clustered into one class, and the clustering processing of the face features of the same face in the face features of the video is realized.

Optionally, the embodiment of the present invention can also extract facial features from the video according to the playing time sequence of the video (especially when the video is a video stream); at this time, the facial features of the video are extracted successively according to the playing time order Therefore, in the embodiment of the present invention, the clustering processing of the facial features of the same face can also be performed according to the sequence of extraction of facial features, that is, for the facial features extracted first, the embodiment of the present invention can perform clustering processing first , for the later extracted facial features, the embodiment of the present invention can perform clustering processing later;

Optionally, FIG. 7 shows a flow chart of a method for clustering facial features of the same face provided by an embodiment of the present invention. The stage of clustering processing of features, such as steps S12 in Figure 3, step S14 in Figure 3, and step S22 in Figure 5; with reference to Figure 7, the method may include:

Step S30, for any face feature to be clustered, detect whether there is a target cluster whose similarity with the face feature to be clustered meets the predetermined similarity requirement among the obtained clusters, if so, execute step S31 , if not, go to step S32.

In the embodiment of the present invention, the face features can be represented by face feature vectors; since the face features are extracted from the video in sequence according to the playing time, the processing server may have previously Other facial features similar to this facial feature have been clustered, or may not have been clustered;

Optionally, in this embodiment of the present invention, for any face feature to be clustered (such as the face feature currently extracted from any video, or, in the case of multiple videos, any video in the second cluster Any face feature of the face feature), detect whether there is a target cluster whose similarity with the face feature to be clustered meets the predetermined similarity requirements in the obtained clusters, so as to judge whether the face feature to be clustered belongs to the same Whether other facial features of the face have been clustered.

Step S31. Gather the face features to be clustered into the target clusters.

When it is detected that among the obtained clusters, there is a target cluster whose similarity with the face feature to be clustered meets the predetermined similarity requirements, it can be considered that other people belonging to the same face as the face feature to be clustered The face features have been clustered, and the face features to be clustered can be aggregated into the target cluster.

Step S32, setting a new cluster, and gathering the face features to be clustered into the new cluster.

When it is detected that among the obtained clusters, there is no target cluster whose similarity with the face feature to be clustered meets the predetermined similarity requirements, it can be considered that the face feature to be clustered belongs to a new face. Separate the face features to be clustered into a cluster, that is, set a new cluster and gather the face features to be clustered into the new cluster.

In this embodiment of the present invention, the similarity between human face features can be represented by the distance of the face feature vectors of human face features; optionally, FIG. Another method flow for clustering processing, the flow shown in Figure 8 can be considered as a refinement of the flow shown in Figure 7, referring to Figure 8, the flow can include:

Step S40, for any face feature to be clustered, detect whether there is a cluster obtained within the predetermined vector distance of the face feature vector of the face feature to be clustered, if yes, execute step S41, if not , execute step S43.

Step S41. After adding the face feature to be clustered into the obtained cluster, whether the corresponding radius of the obtained cluster is greater than the radius threshold, if not, execute step S42, and if yes, execute step S43.

Step S42. Determine the obtained cluster as the target cluster, and gather the face features to be clustered into the target cluster.

Step S43, setting a new cluster, and gathering the face features to be clustered into the new cluster.

It can be seen that if within the predetermined vector distance of the face feature vector corresponding to the face feature to be clustered, there is an obtained cluster, and after adding the face feature to be clustered in the obtained cluster, , the radius of the obtained cluster is not greater than the radius threshold, then it can be determined that the obtained cluster is the target cluster, and the facial features to be clustered can be gathered into the target cluster;

And if within the predetermined vector distance of the corresponding face feature vector of the face feature to be clustered, there is no clustering obtained, then it can be determined that the face feature to be clustered belongs to a new face, and a new Clustering, gathering the face features to be clustered into this new cluster;

If within the predetermined vector distance of the face feature vector corresponding to the face features to be clustered, there are already obtained clusters, but after adding the face features to be clustered in the obtained clusters, the obtained If the radius of the clustering is greater than the radius threshold, then it is determined that the face features clustered by the obtained clusters do not belong to the same face as the face features to be clustered, a new cluster can be set, and all The face features to be clustered are gathered into this new cluster.

Further, after the face features to be clustered are gathered into a cluster (may be the obtained target cluster, or a new cluster), the cluster to which the face features to be clustered are gathered can be updated Corresponding radius and centroid; Further, can also set corresponding video information list (as video time point list, video progress list etc.) After the clustering, it is inserted into the video information list corresponding to the cluster where the face features to be clustered are gathered, and the face features to be clustered are included in the video information of the corresponding video.

Correspondingly, for any face feature to be clustered, when detecting whether there is an obtained cluster within the predetermined vector distance of the face feature vector of the face feature to be clustered, it can be judged that the distance from the face feature vector to be clustered is Whether the face feature vector distance of the clustered face feature is the nearest neighbor of ε, whether there is a cluster obtained;

As an example, taking the clustering of the facial features of the same face in a video as an example, the facial features can be extracted according to the video frame sequence of the video, and the extracted facial features become the facial features to be clustered , according to the video time point <T1, T2, ..., Tn> of the video, the facial features <F1, F2, ...,, Fn> of the same face can be extracted;

For any extracted face feature, it can be judged whether there is a cluster in the neighborhood whose distance from the face feature vector of the face feature is ε; if there is a cluster, you can try to adjust the cluster and confirm to add the face After the feature, whether the update of the radius of the cluster exceeds the maximum radius threshold: if it exceeds the maximum radius threshold, set this face feature as a new cluster, and insert the face feature in the video time point list corresponding to the cluster. Otherwise, after adding the face feature to the cluster, update the centroid and radius of the cluster, and at the same time list the corresponding video time points of the cluster, update the video time of the face feature in the video point, the corresponding diagram can be shown in Figure 9.

What needs to be explained here is that clustering is actually to divide a data set into different classes or clusters according to a certain standard (such as distance criterion), so that the similarity of the data in the same cluster is as large as possible, while not in the same cluster. The diversity of data objects in the cluster is also as large as possible. That is, after clustering, the data of the same class are gathered together as much as possible, and the different data are separated as much as possible; in the corresponding diagram of Figure 9 described above, the empirical radius value ε can be set, and a circle is formed with ε as the radius, and the data in the circle It is considered to be similar, and the corresponding centroid of the cluster is the center of the circle. When the data gathered in a cluster changes, the corresponding centroid and radius of the cluster will be updated accordingly.

It should be further noted that if the video is input to the processing server in the form of a video stream, the processing server can distinguish the video streams of multiple videos by establishing a video index chain; At this time, it can be judged whether the video index chain of the video stream exists through the identification of the video stream, and a video index chain can record the clustering of the facial features of different faces gathered in the video (if under multiple videos, then is the form of the second clustering); if so, then for the video stream, the facial features of the video stream can be extracted according to the sequence of video frames (a form of video playback time order), and it is determined that the extracted facial features are in the Corresponding clustering in the video index chain of video (as shown in Fig. 7 or Fig. 8 mode realization);

If not, then the video index chain of this video stream can be created, extract the face features of this video stream according to the video frame sequence, and determine the corresponding clustering ( as shown in Figure 7 or Figure 8).

Optionally, the video index chain can exist in key-value form. For example, the video index chain of a video can be the identification of the video as the primary key, and the values associated with each cluster of facial features of the same face in the video as the primary key.

Optionally, for any of the above-mentioned videos, the face features of the same face are clustered to obtain the clustering stage of the same face feature (step S12 shown in Figure 3, shown in Figure 5 Step S22 and other stages), the embodiment of the present invention can set the clustering of the corresponding facial features of any video, which is the clustering of the facial features of the main faces in the video;

Optionally, for any video, after at least one cluster of facial features of the same face in the video is obtained, the clusters whose appearance times of facial features in the at least one cluster are less than the number threshold can be deleted;

As in the case of multiple videos, when obtaining the corresponding second clustering process of each video, for any video, after obtaining at least one cluster of the facial features of the same face in the video, the Deleting at least one cluster whose number of occurrences of face features in at least one cluster is less than the number threshold, to obtain at least one second cluster corresponding to the video;

For example, in the case of a video, when obtaining the processing of the corresponding first cluster of the video, after obtaining at least one cluster of the facial features of the same face in the video, the number of occurrences of the facial features can be less than the number of times threshold The clusters are deleted to obtain at least one first cluster corresponding to the video.

Optionally, the video information of the face feature in the corresponding video may include: the video time point of the face feature in the corresponding video; optionally, for any video, the embodiment of the present invention can delete the video by the method shown in FIG. 10 In the clustering of facial features, the clustering of the facial features of secondary faces; with reference to Figure 10, the method may include:

Step S50, for any video, after obtaining at least one cluster of facial features of the same face in the video, determine the distribution of video time points corresponding to the facial features of each cluster in the video.

In order to distinguish between the main face and the secondary face, after clustering the face features of the same face in the video for any video, the face features gathered by each cluster at the video time of the corresponding video can be Points to determine the distribution of video time points, that is, to determine the distribution of face features of each cluster at the corresponding video time points in a video.

Step S51 , deleting the clusters whose distribution of video time points does not fall within at least one set duration interval.

Optionally, the embodiment of the present invention can set at least one set time interval, such as n time intervals whose duration is t; thus, for any video, the video time point corresponding to the facial features of each cluster of the video can be judged Distribution, whether it is in more than the n time intervals of t, if not, it can be determined that the face feature of the cluster belongs to the secondary face in the video, and can be deleted, if so, it can be determined that the cluster The face features of the class belong to the main faces in the video and can be preserved.

The above describes the scheme of building a face index for one or more videos without face information marking from different angles, by clustering the face features of the same face in one or more videos Processing to obtain the first cluster suitable for the one or more videos; Index establishment; further, through each first clustering, can gather the face features of the same face in this one or more videos, can reduce the number of faces in the face index storehouse, can greatly reduce the number of faces Data redundancy for face indexing.

After the face index is established, the embodiment of the present invention can support the user to retrieve the video with the face of the target person by inputting the face image of the target person when retrieving the video; as an application example, the present invention The application process of the embodiment can be shown in Figure 11, referring to Figure 11, the process can include:

Step S60, the terminal sends a video retrieval request to the processing server, and the video retrieval request carries a target face image.

When the user needs to search for a video with a target person, he can input the target face image of the target person to the terminal, and the terminal can correspondingly send a video retrieval request to the processing server, and carry the target face image in the video retrieval request;

As an example, the user can import the target face image on the page of the video client loaded on the terminal, and after the user clicks the search button on the page, the terminal can send a video retrieval request carrying the target face image to the processing server through the video client.

Step S61, the processing server extracts the target face features of the target face image.

After the processing server obtains the video retrieval request, it can parse to obtain the target facial image carried in the video retrieval request, and extract facial features of the target facial image (referred to as target facial features).

Step S62, the processing server determines the video information of the corresponding video associated with the target face feature from the face index.

After the processing server establishes the face index, based on the face features gathered by each first cluster in the face index, the associated face features are in the video information of the corresponding video, and the processing server can determine the associated face features of the target. The video information of the corresponding video (including each video corresponding to the target face feature, and the video time point of the target face feature in each corresponding video, etc.).

Optionally, as an implementation, the processing server may perform a k-nearest neighbor query on the centroids of each first cluster in the face index based on the target face feature, obtain the first cluster where the target face feature is located, and determine The video IDs of the videos corresponding to the facial features gathered in the first cluster where the target facial feature is located, and the video time points in each video.

Step S63, the processing server sends the video information to the terminal.

After determining the video information of the corresponding video associated with the target face feature, the processing server may output the determined video information, that is, send the video information to the terminal.

Step S64, the terminal displays the video information.

Optionally, the terminal may display the video information on the search result page of the video client, such as displaying the video corresponding to the target face feature, and the video time point of the corresponding video.

The application example given above is explained by using a video client to perform video retrieval by the user. Of course, the user can also use a browsing component such as a browser to log in to a video website and perform video retrieval through the video website.

The device for establishing the face index provided by the embodiment of the present invention is introduced below; the device for establishing the face index described below can be considered as the program that the processing server needs to set for realizing the method for establishing the face index provided by the embodiment of the present invention module. The content of the apparatus for establishing a face index described below may be referred to in correspondence with the content of the method for establishing a face index described above.

Fig. 12 is a structural block diagram of a device for establishing a face index provided by an embodiment of the present invention. The device can be applied to a processing server. Referring to Fig. 12, the device may include:

A video acquisition module 100, configured to acquire at least one video;

Facial features and video information determining module 200, for determining respectively the corresponding facial features of each video, and determining the video information of each facial feature in the corresponding video;

The first clustering obtaining module 300 is used to cluster the facial features of the same face to obtain at least one first cluster; wherein, the facial features gathered by a first cluster represent, and the at least one Facial features of the same face in the video;

The face index establishment module 400 is used for associating the video information of the face feature in the corresponding video with respect to each face feature gathered by each first cluster, so as to obtain the face index of the at least one video.

Optionally, in the case of multiple videos, the first clustering obtaining module 300 is configured to perform clustering processing on facial features of the same face to obtain at least one first clustering, specifically including:

Obtain the clustering results of each video, wherein the clustering results of a video at least include: at least one second cluster that gathers the facial features of the same face in the video, and each person gathered by each second cluster The video information of the face feature in the video;

For the clustering results of each video, the facial features of the same face are clustered to obtain at least one first cluster corresponding to the plurality of videos.

Optionally, in the case of multiple videos, the first clustering module 300 is used to obtain the clustering results of each video, specifically including:

For the corresponding facial features of each video, the facial features of the same face are clustered to obtain at least one second cluster corresponding to each video;

For the face features collected by the second clusters corresponding to each video, correlate the face features with the video information of the corresponding video, and obtain the clustering results of each video.

Optionally, in the case of multiple videos, the first cluster obtaining module 300 is configured to obtain at least one second cluster corresponding to each video, specifically including:

For any video, after obtaining at least one cluster of the facial features of the same face in the video, delete the clusters whose facial features in the at least one cluster appear less than the number of times threshold, and obtain the corresponding video At least one second cluster.

Optionally, the video information of the corresponding video of the facial feature at least includes: the video time point of the corresponding video of the facial feature; in the case of multiple videos, the first clustering module 300 is used for any one video, after obtaining at least one cluster of the face features of the same face in the video, delete the clusters whose appearance times of the face features in the at least one cluster are less than the number threshold, and obtain at least one corresponding first cluster of the video. Binary clustering, specifically including:

For any piece of video, after obtaining at least one cluster of the facial features of the same face in the video, determine the corresponding video time point distribution of the facial features of each cluster of the video;

Deleting the clusters whose video time point distribution is not at least one set duration interval, and obtaining at least one second cluster corresponding to the video.

Optionally, in the case of multiple videos, the first clustering module 300 is used to cluster the facial features of the same face for the corresponding facial features of each video, specifically including:

If there is a video index chain in any video obtained, for the face feature extracted from the video, determine the corresponding second clustering of the face feature in the video index chain of the video; wherein, a video index chain records the each second clustering of facial features of different faces clustered in the video;

If there is no video index chain for any video obtained, create a video index chain corresponding to the video, and for the facial features extracted from the video, make sure that the extracted facial features correspond to the created video index chain the second cluster of .

Optionally, for any face feature to be clustered, perform the process of clustering the face features of the same face, specifically including:

Detecting whether there is a target cluster whose similarity with the face feature to be clustered meets the predetermined similarity requirement among the obtained clusters;

If the target cluster exists, gathering the face features to be clustered into the target cluster;

If the target cluster does not exist, a new cluster is set, and the face features to be clustered are gathered into the new cluster.

Optionally, the above-mentioned process of detecting whether there is a target cluster whose similarity with the face feature to be clustered meets the predetermined similarity requirement among the obtained clusters may specifically include:

Detecting whether there is an obtained cluster within a predetermined vector distance of the face feature vector of the face features to be clustered;

If there are clusters that have been obtained, and after the face features to be clustered are added to the clusters that have been obtained, the corresponding radius of the clusters that have been obtained is not greater than the radius threshold, then determine that the clusters that have been obtained clustering for said target;

If there is no obtained cluster, or, there is an obtained cluster, but after adding the face features to be clustered in the obtained cluster, the radius of the obtained cluster is greater than the radius threshold, Then it is determined that the target cluster does not exist.

Optionally, for the facial features to be clustered, performing the process of associating the video information of the corresponding video with the facial features may specifically include:

Insert in the video information list corresponding to the cluster where the face features to be clustered are gathered, and the face features to be clustered are in the video information of the corresponding video.

Optionally, the apparatus for establishing a face index provided in the embodiment of the present invention may also be used to: update the corresponding radius and centroid of the clusters into which the face features to be clustered are gathered.

Optionally, in the embodiment of the present invention, the video information of the face feature in the corresponding video may specifically include:

Facial features at the video time point of the corresponding video, and/or, the video progress, and/or, the identification of the video.

Optionally, the device for establishing a face index provided in the embodiment of the present invention can also be used for:

Obtaining a video retrieval request, the video retrieval request carrying a target face image;

Extracting the target face feature of the target face image;

Determine from the face index, the video information of the corresponding video associated with the target face feature;

The determined video information is output.

The embodiment of the present invention also provides a processing server, which can be loaded with the above-mentioned program module to execute the method for establishing a face index provided by the embodiment of the present invention; the above-mentioned program module can be implemented in the form of a program, as shown in FIG. 13 Referring to FIG. 13 for a structural block diagram of a processing server provided in an embodiment of the present invention, the processing server may include:

At least one processing chip 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;

In the embodiment of the present invention, the number of the processing chip 1, the communication interface 2, the memory 3, and the communication bus 4 is at least one, and the processing chip 1, the communication interface 2, and the memory 3 complete the mutual communication through the communication bus 4;

The processing chip 1 may be a central processing unit CPU, or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention.

The memory 3 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one magnetic disk memory.

Wherein, the memory 3 stores programs, and the processing chip 1 invokes the programs stored in the memory 3 to implement the above-mentioned steps of the method for establishing a face index executed by the processing server.

The program is specifically intended for:

Get at least one video;

Respectively determine the corresponding facial features of each video, and determine the video information of each facial feature in the corresponding video;

The facial features of the same face are clustered to obtain at least one first cluster; wherein, the facial features gathered by a first cluster represent the facial features of the same face in the at least one video;

For each face feature gathered by each first cluster, associate the face feature with the video information of the corresponding video, and obtain the face index of the at least one video.

Optionally, the embodiment of the present invention further provides a storage medium, which can store a program suitable for execution by the processing chip, so as to implement the above-mentioned steps of the method for establishing a face index executed by the processing server.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other. As for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and for relevant details, please refer to the description of the method part.

Professionals can further realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software or a combination of the two. In order to clearly illustrate the possible For interchangeability, in the above description, the composition and steps of each example have been generally described according to their functions. Whether these functions are executed by hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may use different methods to implement the described functions for each specific application, but such implementation should not be regarded as exceeding the scope of the present invention.

The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be directly implemented by hardware, software modules executed by a processing chip, or a combination of both. Software modules can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other Any other known storage medium.

The above description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention will not be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (12)

1. A method for setting up a face index, characterized in that, comprising: Get multiple videos; Respectively determine the corresponding facial features of each video, and determine the video information of each facial feature in the corresponding video; For the corresponding facial features of each video, the facial features of the same face are clustered to obtain at least one second cluster corresponding to each video, specifically including: if any of the acquired videos has video Index chain, for the face feature extracted from the video, determine the corresponding second clustering of the face feature in the video index chain of the video; wherein, a video index chain records the different faces gathered in the video For each second clustering of face features, a video index chain uses the video identity as the primary key, and the values associated with the clusters of the same face’s face features in the video as the primary key; if any video obtained is not There is a video index chain, create the video index chain corresponding to the video, and for the face feature extracted from the video, determine the corresponding second clustering of the extracted face feature in the created video index chain of the video; Respectively for the face features gathered by the corresponding second clusters of each video, correlating the video information of the corresponding video to obtain the clustering results of each video, wherein the clustering result of a video at least includes: Aggregating at least one second clustering of the facial features of the same face in the video, and the video information of each facial feature gathered by each second clustering in the video; a second clustering of a video represents the video A clustering of facial features of the same face in For the second clustering of each video, the face features of the same face are clustered to obtain at least one first cluster corresponding to the plurality of videos; wherein, the faces gathered by a first cluster Feature representation, the face features of the same face in the multiple videos; For each face feature gathered by each first cluster, associate the face feature in the video information of the corresponding video, so as to obtain the face indexes of the multiple videos when no face information is marked, and use the Write the face index of the above multiple videos into the face index database. 2. the method for setting up human face index according to claim 1, is characterized in that, described respectively obtaining at least one second cluster corresponding to each video comprises: For any video, after obtaining at least one cluster of the facial features of the same face in the video, delete the clusters whose facial features in the at least one cluster appear less than the number of times threshold, and obtain the corresponding video At least one second cluster. 3. according to the method for setting up human face index described in any one of claim 1-2, it is characterized in that, for any human face feature to be clustered, carry out described clustering process with the human face feature of same people's face The process includes: Detecting whether there is a target cluster whose similarity with the face feature to be clustered meets the predetermined similarity requirement among the obtained clusters; If the target cluster exists, gathering the face features to be clustered into the target cluster; If the target cluster does not exist, a new cluster is set, and the face features to be clustered are gathered into the new cluster. 4. The method for setting up a face index according to claim 3, characterized in that, in the obtained clusters of the detection, whether there is a similarity with the face features to be clustered that meets the predetermined similarity requirements Target clusters include: Detecting whether there is an obtained cluster within a predetermined vector distance of the face feature vector of the face features to be clustered; If there are already obtained clusters, and after adding the face features to be clustered in the obtained clusters, the corresponding radius of the obtained clusters is not greater than the radius threshold, then determine the obtained clusters class is the target cluster; If there is no obtained cluster, or, there is an obtained cluster, but after adding the face features to be clustered in the obtained cluster, the radius of the obtained cluster is greater than the radius threshold, Then it is determined that the target cluster does not exist. 5. the method for setting up face index according to claim 3, is characterized in that, for described face features to be clustered, the process of performing the associated face features in the video information of corresponding video comprises: Insert in the video information list corresponding to the cluster where the face features to be clustered are gathered, and the face features to be clustered are in the video information of the corresponding video. 6. the method for setting up face index according to claim 3, is characterized in that, described method also comprises: Update the corresponding radius and centroid of the clusters where the face features to be clustered are gathered. 7. the method for setting up face index according to claim 1, is characterized in that, described face feature comprises in the video information of corresponding video: Facial features at the video time point of the corresponding video, and/or, the video progress, and/or, the identification of the video. 8. the method for setting up face index according to claim 2, is characterized in that, the video information of described face feature at corresponding video at least comprises: the video time point of face feature at corresponding video; For any video, after obtaining at least one cluster of the facial features of the same face in the video, delete the clusters whose facial features appear in the at least one cluster less than the number of times threshold, and obtain the video The corresponding at least one second cluster includes: For any piece of video, after obtaining at least one cluster of the facial features of the same face in the video, determine the corresponding video time point distribution of the facial features of each cluster of the video; Deleting the clusters whose video time point distribution is not at least one set duration interval, and obtaining at least one second cluster corresponding to the video. 9. the method for setting up face index according to claim 1, is characterized in that, also comprises: Obtaining a video retrieval request, the video retrieval request carrying a target face image; Extracting the target face feature of the target face image; Determine from the face index, the video information of the corresponding video associated with the target face feature; The determined video information is output. 10. A device for establishing a face index, comprising: A video acquisition module, configured to acquire multiple videos; Facial features and video information determining module, used to respectively determine the corresponding facial features of each video, and determine the video information of each facial feature in the corresponding video; The first clustering obtaining module is used to cluster the face features of the same face for the corresponding face features of each video, so as to obtain at least one second cluster corresponding to each video, specifically including : If there is a video index chain in any acquired video, for the face feature extracted from the video, determine the corresponding second cluster of the face feature in the video index chain of the video; wherein, a video index chain record Each of the second clusters of the facial features of different faces gathered in the video, a video index chain with the identity of the video as the primary key, and the clusters associated with the facial features of the same face in the video as the primary key value; if there is no video index chain for any acquired video, create the corresponding video index chain for the video, and for the facial features extracted from the video, make sure that the extracted facial features are in the created video index chain for the video Corresponding second clustering in each video; Respectively aiming at the facial features gathered by each second clustering corresponding to each video, correlating facial features in the video information of the corresponding video to obtain the clustering results of each video, wherein one The clustering result of the video at least includes: at least one second cluster that aggregates the facial features of the same face in the video, and the video information of each facial feature gathered by each second cluster in the video; A second clustering represents a clustering of the facial features of the same face in the video; for the second clustering of each video, the facial features of the same face are clustered to obtain the multiple videos Corresponding to at least one first cluster; wherein, the face features gathered by a first cluster represent the face features of the same face in the multiple videos; The face index building module is used to associate face features with the video information of the corresponding video for each face feature gathered by each first cluster, so as to obtain the plurality of videos without any face information mark The face index of the video, and the face index of the multiple videos is written into the face index storehouse. 11. A processing server, characterized in that it comprises: at least one memory and at least one processing chip, the memory stores a program, and the processing chip invokes the program to implement any one of claims 1-9 The steps of the method for establishing a face index. 12. A storage medium, characterized in that the storage medium stores a program suitable for execution by a processing chip to implement the steps of the method for establishing a face index according to any one of claims 1-9.

Images