CN111401326A - Target identity recognition method based on picture recognition, server and storage medium - Google Patents
Target identity recognition method based on picture recognition, server and storage medium Download PDFInfo
- Publication number
- CN111401326A CN111401326A CN202010316554.8A CN202010316554A CN111401326A CN 111401326 A CN111401326 A CN 111401326A CN 202010316554 A CN202010316554 A CN 202010316554A CN 111401326 A CN111401326 A CN 111401326A
- Authority
- CN
- China
- Prior art keywords
- image
- target
- region
- similarity value
- target object
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/20—Scenes; Scene-specific elements in augmented reality scenes
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a target identity recognition method based on picture recognition, which is applied to a server, the method comprises inputting a first image containing a first target object into a target recognition model to output a second image, inputting a second image into a feature extraction model to output a first image feature, recognizing a first region image contained in the second image by the first image feature input region recognition model, acquiring attribute information of the first region image, inquiring a plurality of second region images identical to the attribute information from a database, respectively calculating similarity values between the first region image and each second region image, and calculating comprehensive similarity values of the first target object and each second target object according to the similarity value and the weight value of each first region image, and selecting the identity information of the second target object corresponding to the maximum comprehensive similarity value as the identity information of the first target object. The invention can improve the accuracy of identifying the identity information of the target object.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a target identity recognition method based on picture recognition, a server and a storage medium.
Background
During the moving and loading of goods, the identity information of the goods, such as containers, needs to be determined. At present, most goods are intelligently acquired through a camera and an OCR recognition technology.
The OCR recognition method has the following problems: the method has high requirement on the relative position of the goods and the camera, namely for a single goods, the camera needs to be capable of acquiring images on the front side or close to the front side, and when the shooting angle exists between the camera and the goods, so that image data of a plurality of sides simultaneously exist in a shot picture, the image distortion is easily generated, so that the recognition accuracy of the OCR technology is influenced, and therefore how to improve the accuracy of the identification information of the target object becomes a technical problem which needs to be solved urgently.
Disclosure of Invention
The invention mainly aims to provide a target identity recognition method based on image recognition, a server and a storage medium, aiming at solving the problem of how to improve the accuracy of target identity information recognition.
In order to achieve the above object, the present invention provides a target identity recognition method based on image recognition, which is applied to a server, and the method includes:
a receiving step: acquiring a first image of a first target object containing identity information to be recognized, inputting the first image into a pre-trained target recognition model, and outputting a second image;
an identification step: inputting the second image into a pre-trained feature extraction model, outputting a first image feature, inputting the first image feature into a pre-trained region identification model, identifying a first region image contained in the second image, and assigning a unique weight value to each type of the first region image;
a calculation step: inquiring a plurality of second area images matched with the attribute information of the first area image from a database of second area images of a second target object in which the identification information is stored in advance, and respectively calculating the similarity value between the first area image and each matched second area image; and
a confirmation step: and calculating the comprehensive similarity value of the first target object and each second target object according to the similarity value and the weight value of each first region image, and selecting the identity information of the second target object corresponding to the image with the largest comprehensive similarity value as the identity information of the first target object.
Preferably, the training process of the target recognition model is as follows:
acquiring first image samples, and marking the area of the first target object in each first image sample;
dividing the first image sample into a training set and a verification set according to a preset proportion, wherein the number of the image samples in the training set is greater than that of the image samples in the verification set;
inputting the image samples in the training set into the target recognition model for training, verifying the target recognition model by using the verification set every other preset period, and verifying the accuracy of the target recognition model by using each first image sample in the verification set and the corresponding first target object; and
and when the verification accuracy is greater than a preset threshold value, finishing the training to obtain the trained target recognition model.
Preferably, the similarity value algorithm includes:
wherein x isiImage features, y, representing the first region imageiRepresenting image features of the second region image, d (x, y) representing xiAnd yiA similarity value between the first region image and the second region image.
Preferably, the comprehensive similarity value between the first target object and each second target object is calculated according to a preset calculation rule, and the calculation rule includes a first calculation rule and a second calculation rule:
when the second image comprises a first area image with attribute information as a preset attribute, calculating a comprehensive similarity value of the first area image by adopting the first calculation rule, and taking the identity information of a second target corresponding to the comprehensive similarity value as the identity information of the first target;
and when the second image does not contain the first preset information, calculating a comprehensive similarity value of second preset information by adopting the second calculation rule, and taking the identity information of a second target corresponding to the comprehensive similarity value as the identity information of the first target.
Preferably, the first calculation rule is:
wherein d represents a combined similarity value of the first and second objects, d1A similarity value between a first region image and a second region image representing that the attribute information is a preset attribute, wherein lambda represents a weight value of the first region image, and lambda represents a weight value of the first region image>0。
Preferably, the second calculation rule is:
wherein d represents a combined similarity value of the first and second objects, diRepresenting the space between the first and second region imagesM represents the number of the first region images, m is less than or equal to 3, i represents the sequence mark of the first region images, λ represents the weight value of the first region images, and λ>0。
To achieve the above object, the present invention further provides a server, including a memory and a processor, where the memory stores a target exception identifying program, and the target exception identifying program, when executed by the processor, implements the following steps:
a receiving step: acquiring a first image of a first target object containing identity information to be recognized, inputting the first image into a pre-trained target recognition model, and outputting a second image;
an identification step: inputting the second image into a pre-trained feature extraction model, outputting a first image feature, inputting the first image feature into a pre-trained region identification model, identifying a first region image contained in the second image, and assigning a unique weight value to each type of the first region image;
a calculation step: inquiring a plurality of second area images matched with the attribute information of the first area image from a database of second area images of a second target object in which the identification information is stored in advance, and respectively calculating the similarity value between the first area image and each matched second area image; and
a confirmation step: and calculating the comprehensive similarity value of the first target object and each second target object according to the similarity value and the weight value of each first region image, and selecting the identity information of the second target object corresponding to the image with the largest comprehensive similarity value as the identity information of the first target object.
Preferably, the training process of the target recognition model is as follows:
acquiring first image samples, and marking the area of the first target object in each first image sample;
dividing the first image sample into a training set and a verification set according to a preset proportion, wherein the number of the image samples in the training set is greater than that of the image samples in the verification set;
inputting the image samples in the training set into the target recognition model for training, verifying the target recognition model by using the verification set every other preset period, and verifying the accuracy of the target recognition model by using each first image sample in the verification set and the corresponding first target object; and
and when the verification accuracy is greater than a preset threshold value, finishing the training to obtain the trained target recognition model.
Preferably, the similarity value algorithm includes:
wherein x isiImage features, y, representing the first region imageiRepresenting image features of the second region image, d (x, y) representing xiAnd yiA similarity value between the first region image and the second region image.
To achieve the above object, the present invention further provides a computer readable storage medium, which stores thereon a target anomaly identification program, which is executable by one or more processors to implement the steps of the image recognition-based target identity recognition method as described above.
The invention provides a target identity recognition method based on picture recognition, a server and a storage medium, which are characterized in that a first image containing a first target object is obtained, the first image is input into a target recognition model to output a second image, a second image input feature extraction model outputs a first image feature, a first image feature input region recognition model recognizes a first region image contained in the second image, a plurality of second region images of different second target objects are created in a database, attribute information of the first region image is obtained, a plurality of second region images identical to the attribute information are inquired from the database, similarity values between the first region image and the second region images are respectively calculated, comprehensive similarity values of the first target object and the second target objects are calculated according to the similarity values and weight values of the first region images, and identity information of the second target object corresponding to the maximum comprehensive similarity value is selected as the first target object Identity information of (2). The invention can improve the accuracy of identifying the identity information of the target object.
Drawings
FIG. 1 is a diagram of an application environment of a server according to a preferred embodiment of the present invention;
FIG. 2 is a block diagram of a preferred embodiment of the target exception identifier process of FIG. 1;
fig. 3 is a flowchart illustrating a target identity recognition method based on image recognition according to a preferred embodiment of the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
In order to make the objects, technical embodiments and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the description relating to "first", "second", etc. in the present invention is for descriptive purposes only and is not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In addition, the technical embodiments of the present invention may be combined with each other, but it must be based on the realization of those skilled in the art, and when the combination of the technical embodiments contradicts each other or cannot be realized, such combination of the technical embodiments should be considered to be absent and not within the protection scope of the present invention.
The invention provides a server 1.
The server 1 includes, but is not limited to, a memory 11, a processor 12, and a network interface 13.
The memory 11 includes at least one type of readable storage medium, which includes a flash memory, a hard disk, a multimedia card, a card type memory (e.g., SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, and the like. The memory 11 may in some embodiments be an internal storage unit of the server 1, for example a hard disk of the server 1. The memory 11 may also be an external storage device of the server 1 in other embodiments, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like, provided on the server 1.
Further, the memory 11 may also include both an internal storage unit of the server 1 and an external storage device. The memory 11 may be used not only to store application software installed in the server 1 and various types of data such as the code of the target abnormality recognition program 10, but also to temporarily store data that has been output or is to be output.
The network interface 13 may optionally comprise a standard wired interface, a wireless interface (e.g. WI-FI interface), typically used for establishing a communication connection between the server 1 and other electronic devices.
The client can be a desktop computer, a notebook, a tablet computer, a mobile phone, and the like.
Examples of such wired and wireless communication protocols may include, but are not limited to, at least one of transmission control protocol and internet protocol (TCP/IP), User Datagram Protocol (UDP), hypertext transfer protocol (HTTP), File Transfer Protocol (FTP), ZigBee, EDGE, IEEE 802.11, optical fidelity (L i-Fi), 802.16, IEEE 802.11s, IEEE 802.11g, multi-hop communications, wireless Access Points (APs), device-to-device communications, cellular communication protocols, and/or Bluetooth (ToBlueth) communication protocols, or combinations thereof.
Optionally, the server 1 may further comprise a user interface, which may comprise a Display (Display), an input unit such as a Keyboard (Keyboard), and an optional user interface may also comprise a standard wired interface, a wireless interface, optionally, in some embodiments, the Display may be an L ED Display, a liquid crystal Display, a touch-sensitive liquid crystal Display, an O L ED (Organic light-Emitting Diode) touch-sensitive device, and the like, wherein the Display may also be referred to as a Display screen or a Display unit for displaying information processed in the server 1 and for displaying a visualized user interface.
While FIG. 1 shows only a server 1 having components 11-13 and a target anomaly recognition program 10, those skilled in the art will appreciate that the configuration shown in FIG. 1 is not limiting of server 1, and may include fewer or more components than shown, or some components in combination, or a different arrangement of components.
In the present embodiment, the target anomaly recognition program 10 of fig. 1, when executed by the processor 12, implements the following steps:
a receiving step: acquiring a first image of a first target object containing identity information to be recognized, inputting the first image into a pre-trained target recognition model, and outputting a second image;
an identification step: inputting the second image into a pre-trained feature extraction model, outputting a first image feature, inputting the first image feature into a pre-trained region identification model, identifying a first region image contained in the second image, and assigning a unique weight value to each type of the first region image;
a calculation step: inquiring a plurality of second area images matched with the attribute information of the first area image from a database of second area images of a second target object in which the identification information is stored in advance, and respectively calculating the similarity value between the first area image and each matched second area image; and
a confirmation step: and calculating the comprehensive similarity value of the first target object and each second target object according to the similarity value and the weight value of each first region image, and selecting the identity information of the second target object corresponding to the image with the largest comprehensive similarity value as the identity information of the first target object.
For detailed description of the above steps, please refer to the following description of fig. 2 regarding a schematic diagram of program modules of an embodiment of the target anomaly identification program 10 and fig. 3 regarding a schematic diagram of a method flow of an embodiment of a target identification method based on image recognition.
Referring to fig. 2, a schematic diagram of program modules of the target anomaly identification program 10 in fig. 1 is shown. The target abnormality recognition program 10 is divided into a plurality of modules, which are stored in the memory 11 and executed by the processor 12, to complete the present invention. The modules referred to herein are referred to as a series of computer program instruction segments capable of performing specified functions.
In the present embodiment, the target anomaly identification program 10 includes a receiving module 110, an identifying module 120, a calculating module 130 and a confirming module 140.
The receiving module 110 is configured to obtain a first image of a first target object including identity information to be recognized, input the first image into a pre-trained target recognition model, and output a second image.
In the present embodiment, the server 1 receives a first image, such as an image a, uploaded by a client (e.g., a camera or other shooting terminal with a shooting function, or a device with a shooting function and an image transmission function). The first image is a photographed original image, and a first object (such as a container) containing identity information to be recognized and other irrelevant image data, such as a background image of the first object, may exist at the same time. The first image is input to a pre-trained target recognition model, and a second image corresponding to the first image, for example, image B, is output.
In this embodiment, the target recognition model is a Convolutional Neural Network (CNN) model, and the training process of the Convolutional neural network model is as follows:
obtaining first image samples, each of which is marked with a third-party marking tool (e.g., L peeling) to indicate the area of the first object;
dividing the first image sample into a training set and a verification set according to a preset proportion (for example, 5:1), wherein the number of the image samples in the training set is greater than that of the image samples in the verification set;
inputting the image samples in the training set into the convolutional neural network model for training, verifying the convolutional neural network model by using the verification set every preset period (for example, every 1000 times of iteration), and verifying the accuracy of the identification model by using each first image sample and the corresponding first target object in the verification set; and
and when the verification accuracy is greater than a preset threshold (for example, 85%), ending the training to obtain the target recognition model.
The identifying module 120 is configured to input the second image into a pre-trained feature extraction model, output a first image feature, input the first image feature into a pre-trained region identification model, identify a first region image included in the second image, and assign a unique weight value to each type of the first region image.
Because the client possibly has the problem of shooting angle in the shooting process, illumination reflection interference can be caused to the second image, and subsequent data processing is influenced. Therefore, in order to avoid this, in the present embodiment, the second image needs to be preprocessed before being input to the feature extraction model. The preprocessing comprises Gaussian filtering, mean filtering, Gamma correction, histogram equalization and the like.
After the preprocessing operation on the second image is completed, the second image is input into the feature extraction model to obtain a first image feature, such as a first image feature a. In this embodiment, the first image feature extraction model is obtained by training a ResNet18 network model, the ResNet18 network model is a convolutional neural network structure model, and the ResNet18 network model can efficiently and quickly identify an image with low resolution, has the characteristic of small bandwidth occupied by calculation, and can be carried on a mobile device for use. The ResNet18 network model includes 17 convolutional layers and 1 fully-connected layer connected in sequence.
In other embodiments, when training the ResNet18 network model, a loss function may be set for the ResNet18 network model in advance, a training sample is input into the ResNet18 network model, forward propagation is performed on the input training sample to obtain an actual output, a preset target output and the actual output are substituted into the loss function, a loss value of the loss function is calculated, backward propagation is performed, parameters of the ResNet18 network model are optimized by using the loss value, and the optimized ResNet18 network model is obtained. And then selecting a training sample to be input into the optimized ResNet18 network model, and training the optimized ResNet18 network model again by referring to the operation until the condition of stopping training is reached.
After the first image features of the second image are extracted, the first image features are input into a pre-trained region identification model, and a first region image, such as an image C, included in the second image is identified. Each first region image is assigned a unique weight value.
In this embodiment, the region recognition model also adopts a convolutional neural network model, a training process is substantially consistent with the target recognition model, a third-party labeling tool (e.g., L unfolding) needs to be used to label the first image region in the second image in the training process, and other details are not repeated herein.
In this embodiment, the first target object is a container, the second image includes three categories, the first category is an image with very good quality, and includes a pure surface image, a first image without occlusion, with normal illumination and with sufficient text information, and in this case, image features extracted from five surfaces of the whole container can be regarded as surfaces capable of well embodying container identity information. In this embodiment, the first-type second image map means that the percentage of the objects to be recognized is 90% or more. The second type is that a shooting angle exists between the client and the container, but the angle is less than 45 degrees and at least most information of one side of the container can be obtained, and the ratio of the target to be identified is 60-90%. The image is the most extensive data type in practical application, and is also the data type which needs to be processed and identified in the scheme. The third type means that a shooting angle exists between the client and the container, the angle is larger than 45 degrees, and the ratio of the target to be identified is lower than 60 percent. Such images are distorted due to too large angle, most of image feature data are lost, and the images cannot provide enough information in the training of the model, so the images are discarded after being judged to belong to the images.
A calculating module 130, configured to query a plurality of second region images that are matched with the attribute information of the first region image from a database of second region images of a second target object in which identification information is stored in advance, and calculate similarity values between the first region image and each matched second region image.
In the present embodiment, the object is a container, and it is assumed that the container photographed in the first image has both a left side and a front side, and attribute information such as a back side, a left side, a front side, a right side, and a top side is provided according to the first area image. And querying a plurality of second area images which are the same as the attribute information from the database, and respectively calculating similarity values between the first area images and the second area images.
The second area image refers to a second area image of a second object (such as a container) stored in the database in advance, and the identity information of the second object is determined. For example, three containers with identity information (for example, in the form of letters or numbers or a combination of the letters and the numbers) of D, E, F respectively are stored in a database in advance, each container has second area images corresponding to the left side, the front side, the right side, the back side and the top side, after the attribute information of the first area image is determined, the second area image consistent with the attribute information is found from the database, the similarity value between the first area image and each second area image is calculated respectively based on a similarity value calculation formula, and preparation is made for calculating a comprehensive similarity value subsequently.
For example, the first image features a of the left side of the container to be recognized are calculated separatelyLeft side ofAnd D second image characteristic b of left side surface of containerD leftThe similarity value d betweenD2；
First image characteristic a of the left side of a container to be identifiedLeft side ofSecond image characteristic b of left side of E containerE left sideThe similarity value d betweenE2；
First image characteristic a of the left side of a container to be identifiedLeft side ofSecond image characteristic b of left side of F containerF leftThe similarity value d betweenE3；
First image characteristic a of the front side of a container to be identifiedIs justSecond image characteristic b of front surface of D containerD is justThe similarity value d betweenD3；
First image characteristic a of the front side of a container to be identifiedIs justSecond image characteristic b of front face of E containerE left sideThe similarity value d betweenE3；
First image characteristic a of the front side of a container to be identifiedIs justSecond image characteristic b of front face of F containerF leftThe similarity value d betweenF3。
The similarity value calculation mode adopts an Euclidean distance algorithm, and the Euclidean distance algorithm is as follows:
wherein x isiImage features, y, representing the first region imageiRepresenting image features of the second region image, d (x, y) representing xiAnd yiA similarity value between the first region image and the second region image.
The confirming module 140 is configured to calculate a comprehensive similarity value between the first object and each second object according to the similarity value and the weight value of each first region image, and select the identity information of the second object corresponding to the maximum comprehensive similarity value as the identity information of the first object.
The target object is a container as an example, and the relative position of the container and the camera is high in the prior art. In other words, for a single container, the camera needs to be directly opposite to the side marked with the container identity information, and in practical application, once a large angle is generated between the camera and the container, a shot image is distorted, so that the recognition rate of the OCR technology on the container identity information is reduced. Therefore, in this embodiment, the comprehensive similarity value between the first object and each of the second objects is calculated, and the identity information of the second object corresponding to the object with the largest comprehensive similarity value is selected as the identity information of the first object.
Specifically, after the similarity value corresponding to each first region image is obtained through calculation, the comprehensive similarity value between the first target object and the second target object in the database is calculated respectively based on a preset calculation rule. Taking the above example as an example, three sets of data of the comprehensive similarity values are obtained, and the identity information of the second target corresponding to the one with the largest similarity value is obtained from the database as the identity information of the first target. And assuming that the comprehensive similarity value between the container to be identified and the D container is the maximum, the identity information of the container to be identified is D.
The scheme can also be applied to the condition that the accuracy of the identification result is not high due to the fact that the individual image area of the shot first image is shielded or the light is insufficient. By introducing the integrated similarity, the identity information of the first object can be determined by integrating the plurality of first image regions.
Further, the comprehensive similarity value of the first target object and each second target object is obtained by calculation according to a preset calculation rule, wherein the calculation rule comprises a first calculation rule and a second calculation rule. When the second image comprises a first area image with attribute information as a preset attribute, calculating a comprehensive similarity value of the first area image by adopting the first calculation rule, and taking the identity information of a second target corresponding to the comprehensive similarity value as the identity information of the first target;
and when the second image does not contain the first preset information, calculating a comprehensive similarity value of second preset information by adopting the second calculation rule, and taking the identity information of a second target corresponding to the comprehensive similarity value as the identity information of the first target.
The object is taken as a container as an example, because the identity information of the container is mainly reflected on the back of the container under normal conditions, the importance degree of the back in calculating the comprehensive similarity value needs to be considered in a focused manner when the back appears in the first image, so that the accuracy of determining the identity information of the first object is improved. Therefore, in this embodiment, the calculation rule includes a first calculation rule and a second calculation rule, and different calculation rules are adopted to calculate the comprehensive similarity value of the first target object according to the actual situation.
When the second image comprises a first area image with attribute information as preset attributes, calculating a comprehensive similarity value of the first area image by adopting the first calculation rule, and taking the identity information of a second target corresponding to the comprehensive similarity value as the identity information of the first target;
and when the second image does not contain the first preset information, calculating a comprehensive similarity value of second preset information by adopting the second calculation rule, and taking the identity information of a second target corresponding to the comprehensive similarity value as the identity information of the first target.
Further, the first calculation rule is:
wherein d represents a combined similarity value of the first and second objects, d1Representing said propertySimilarity value between the first area image and the second area image with information of preset attributes, wherein lambda represents the weight value of the first area image, and lambda>0。
The second calculation rule is:
wherein d represents a combined similarity value of the first and second objects, diRepresenting the similarity value between the first area image and the second area image, m representing the number of the first area images and being less than or equal to 3, i representing the sequence mark of the first area image, λ representing the weight value of the first area image, and λ>0。
In addition, the invention also provides a target identity recognition method based on image recognition. Fig. 3 is a schematic method flow diagram of an embodiment of the image recognition-based target identity recognition method according to the present invention. When the processor 12 of the server 1 executes the target abnormality recognition program 10 stored in the memory 11, the following steps of the target identification method based on picture recognition are realized:
s110, acquiring a first image of a first target object containing identity information to be recognized, inputting the first image into a pre-trained target recognition model, and outputting a second image.
In the present embodiment, the server 1 receives a first image, such as an image a, uploaded by a client (e.g., a camera or other shooting terminal with a shooting function, or a device with a shooting function and an image transmission function). The first image is a photographed original image, and a first object (such as a container) containing identity information to be recognized and other irrelevant image data, such as a background image of the first object, may exist at the same time. The first image is input to a pre-trained target recognition model, and a second image corresponding to the first image, for example, image B, is output.
In this embodiment, the target recognition model is a Convolutional Neural Network (CNN) model, and the training process of the Convolutional neural network model is as follows:
obtaining first image samples, each of which is marked with a third-party marking tool (e.g., L peeling) to indicate the area of the first object;
dividing the first image sample into a training set and a verification set according to a preset proportion (for example, 5:1), wherein the number of the image samples in the training set is greater than that of the image samples in the verification set;
inputting the image samples in the training set into the convolutional neural network model for training, verifying the convolutional neural network model by using the verification set every preset period (for example, every 1000 times of iteration), and verifying the accuracy of the identification model by using each first image sample and the corresponding first target object in the verification set; and
and when the verification accuracy is greater than a preset threshold (for example, 85%), ending the training to obtain the target recognition model.
And S120, inputting the second image into a pre-trained feature extraction model, outputting a first image feature, inputting the first image feature into a pre-trained region identification model, identifying a first region image contained in the second image, and assigning a unique weight value to each type of the first region image.
Because the client possibly has the problem of shooting angle in the shooting process, illumination reflection interference can be caused to the second image, and subsequent data processing is influenced. Therefore, in order to avoid this, in the present embodiment, the second image needs to be preprocessed before being input to the feature extraction model. The preprocessing comprises Gaussian filtering, mean filtering, Gamma correction, histogram equalization and the like.
After the preprocessing operation on the second image is completed, the second image is input into the feature extraction model to obtain a first image feature, such as a first image feature a. In this embodiment, the first image feature extraction model is obtained by training a ResNet18 network model, the ResNet18 network model is a convolutional neural network structure model, and the ResNet18 network model can efficiently and quickly identify an image with low resolution, has the characteristic of small bandwidth occupied by calculation, and can be carried on a mobile device for use. The ResNet18 network model includes 17 convolutional layers and 1 fully-connected layer connected in sequence.
In other embodiments, when training the ResNet18 network model, a loss function may be set for the ResNet18 network model in advance, a training sample is input into the ResNet18 network model, forward propagation is performed on the input training sample to obtain an actual output, a preset target output and the actual output are substituted into the loss function, a loss value of the loss function is calculated, back propagation is performed, parameters of the ResNet18 network model are optimized by using the loss value, and the optimized ResNet18 network model is obtained. And then selecting a training sample to be input into the optimized ResNet18 network model, and training the optimized ResNet18 network model again by referring to the operation until the condition of stopping training is reached.
After the first image features of the second image are extracted, the first image features are input into a pre-trained region identification model, and a first region image, such as an image C, included in the second image is identified. Each first region image is assigned a unique weight value.
In this embodiment, the region recognition model also adopts a convolutional neural network model, a training process is substantially consistent with the target recognition model, a third-party labeling tool (e.g., L unfolding) needs to be used to label the first image region in the second image in the training process, and other details are not repeated herein.
In this embodiment, the first target object is a container, the second image includes three categories, the first category is an image with very good quality, and includes a pure surface image, a first image without occlusion, with normal illumination and with sufficient text information, and in this case, image features extracted from five surfaces of the whole container can be regarded as surfaces capable of well embodying container identity information. In this embodiment, the first-type second image map means that the percentage of the objects to be recognized is 90% or more. The second type is that a shooting angle exists between the client and the container, but the angle is less than 45 degrees and at least most information of one side of the container can be obtained, and the ratio of the target to be identified is 60-90%. The image is the most extensive data type in practical application, and is also the data type which needs to be processed and identified in the scheme. The third type means that a shooting angle exists between the client and the container, the angle is larger than 45 degrees, and the ratio of the target to be identified is lower than 60 percent. Such images are distorted due to too large angle, most of image feature data are lost, and the images cannot provide enough information in the training of the model, so the images are discarded after being judged to belong to the images.
S130, searching a plurality of second region images matched with the attribute information of the first region image in a database of second region images of a second target object in which identification information is stored in advance, and calculating similarity values between the first region image and each of the matched second region images.
In the present embodiment, the object is a container, and it is assumed that the container photographed in the first image has both a left side and a front side, and attribute information such as a back side, a left side, a front side, a right side, and a top side is provided according to the first area image. And querying a plurality of second area images which are the same as the attribute information from the database, and respectively calculating similarity values between the first area images and the second area images.
The second area image refers to a second area image of a second object (such as a container) stored in the database in advance, and the identity information of the second object is determined. For example, three containers with identity information (for example, in the form of letters or numbers or a combination of the letters and the numbers) of D, E, F respectively are stored in a database in advance, each container has second area images corresponding to the left side, the front side, the right side, the back side and the top side, after the attribute information of the first area image is determined, the second area image consistent with the attribute information is found from the database, the similarity value between the first area image and each second area image is calculated respectively based on a similarity value calculation formula, and preparation is made for calculating a comprehensive similarity value subsequently.
For example, the first image features a of the left side of the container to be recognized are calculated separatelyLeft side ofAnd D second image characteristic b of left side surface of containerD leftThe similarity value d betweenD2；
First image characteristic a of the left side of a container to be identifiedLeft side ofSecond image characteristic b of left side of E containerE left sideThe similarity value d betweenE2；
First image characteristic a of the left side of a container to be identifiedLeft side ofSecond image characteristic b of left side of F containerF leftThe similarity value d betweenE3；
First image characteristic a of the front side of a container to be identifiedIs justSecond image characteristic b of front surface of D containerD is justThe similarity value d betweenD3；
First image characteristic a of the front side of a container to be identifiedIs justSecond image characteristic b of front face of E containerE left sideThe similarity value d betweenE3；
First image characteristic a of the front side of a container to be identifiedIs justSecond image characteristic b of front face of F containerF leftThe similarity value d betweenF3。
The similarity value calculation mode adopts an Euclidean distance algorithm, and the Euclidean distance algorithm is as follows:
wherein x isiImage features, y, representing the first region imageiRepresenting image features of the second region image, d (x, y) representing xiAnd yiA similarity value between the first region image and the second region image.
And S140, calculating a comprehensive similarity value between the first target object and each second target object according to the similarity value and the weight value of each first region image, and selecting the identity information of the second target object corresponding to the maximum comprehensive similarity value as the identity information of the first target object.
The target object is a container as an example, and the relative position of the container and the camera is high in the prior art. In other words, for a single container, the camera needs to be directly opposite to the side marked with the container identity information, and in practical application, once a large angle is generated between the camera and the container, a shot image is distorted, so that the recognition rate of the OCR technology on the container identity information is reduced. Therefore, in this embodiment, the comprehensive similarity value between the first object and each of the second objects is calculated, and the identity information of the second object corresponding to the object with the largest comprehensive similarity value is selected as the identity information of the first object.
Specifically, after the similarity value corresponding to each first region image is obtained through calculation, the comprehensive similarity value between the first target object and the second target object in the database is calculated respectively based on a preset calculation rule. Taking the above example as an example, three sets of data of the comprehensive similarity values are obtained, and the identity information of the second target corresponding to the one with the largest similarity value is obtained from the database as the identity information of the first target. And assuming that the comprehensive similarity value between the container to be identified and the D container is the maximum, the identity information of the container to be identified is D.
The scheme can also be applied to the condition that the accuracy of the identification result is not high due to the fact that the individual image area of the shot first image is shielded or the light is insufficient. By introducing the integrated similarity, the identity information of the first object can be determined by integrating the plurality of first image regions.
Further, the comprehensive similarity value of the first target object and each second target object is obtained by calculation according to a preset calculation rule, wherein the calculation rule comprises a first calculation rule and a second calculation rule. When the second image comprises a first area image with attribute information as a preset attribute, calculating a comprehensive similarity value of the first area image by adopting the first calculation rule, and taking the identity information of a second target corresponding to the comprehensive similarity value as the identity information of the first target;
and when the second image does not contain the first preset information, calculating a comprehensive similarity value of second preset information by adopting the second calculation rule, and taking the identity information of a second target corresponding to the comprehensive similarity value as the identity information of the first target.
The object is taken as a container as an example, because the identity information of the container is mainly reflected on the back of the container under normal conditions, the importance degree of the back in calculating the comprehensive similarity value needs to be considered in a focused manner when the back appears in the first image, so that the accuracy of determining the identity information of the first object is improved. Therefore, in this embodiment, the calculation rule includes a first calculation rule and a second calculation rule, and different calculation rules are adopted to calculate the comprehensive similarity value of the first target object according to the actual situation.
When the second image comprises a first area image with attribute information as preset attributes, calculating a comprehensive similarity value of the first area image by adopting the first calculation rule, and taking the identity information of a second target corresponding to the comprehensive similarity value as the identity information of the first target;
and when the second image does not contain the first preset information, calculating a comprehensive similarity value of second preset information by adopting the second calculation rule, and taking the identity information of a second target corresponding to the comprehensive similarity value as the identity information of the first target.
Further, the first calculation rule is:
wherein d represents a combined similarity value of the first and second objects, d1A similarity value between a first region image and a second region image representing that the attribute information is a preset attribute, wherein lambda represents a weight value of the first region image, and lambda represents a weight value of the first region image>0。
The second calculation rule is:
wherein d represents a combined similarity value of the first and second objects, diRepresenting the similarity value between the first area image and the second area image, m representing the number of the first area images and being less than or equal to 3, i representing the sequence mark of the first area image, λ representing the weight value of the first area image, and λ>0。
In addition, the embodiment of the present invention further provides a computer-readable storage medium, which may be any one of or any combination of a hard disk, a multimedia card, an SD card, a flash memory card, an SMC, a Read Only Memory (ROM), an Erasable Programmable Read Only Memory (EPROM), a portable compact disc read only memory (CD-ROM), a USB memory, and the like. The computer-readable storage medium includes a target anomaly identification program 10, and the specific implementation of the computer-readable storage medium of the present invention is substantially the same as the above-mentioned target identity identification method based on image identification and the specific implementation of the server 1, and will not be described herein again.
It should be noted that the sequence of the above embodiments of the present invention is only for description, and does not represent the merits of the embodiments. And the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, apparatus, article, or method that includes the element.
The above description of the embodiments of the present invention is for illustrative purposes only and does not represent the merits of the embodiments. Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical embodiments of the present invention may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) as described above and includes instructions for enabling a terminal device (such as a mobile phone, a computer, a server, or a network device) to execute the method according to the embodiments of the present invention.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (10)
1. A target identity recognition method based on image recognition is applied to a server, and is characterized by comprising the following steps:
a receiving step: acquiring a first image of a first target object containing identity information to be recognized, inputting the first image into a pre-trained target recognition model, and outputting a second image;
an identification step: inputting the second image into a pre-trained feature extraction model, outputting a first image feature, inputting the first image feature into a pre-trained region identification model, identifying a first region image contained in the second image, and assigning a unique weight value to each type of the first region image;
a calculation step: inquiring a plurality of second area images matched with the attribute information of the first area image from a database of second area images of a second target object in which the identification information is stored in advance, and respectively calculating the similarity value between the first area image and each matched second area image; and
a confirmation step: and calculating the comprehensive similarity value of the first target object and each second target object according to the similarity value and the weight value of each first region image, and selecting the identity information of the second target object corresponding to the image with the largest comprehensive similarity value as the identity information of the first target object.
2. The method for identifying the target based on the picture recognition as claimed in claim 1, wherein the training process of the target recognition model is as follows:
acquiring first image samples, and marking the area of the first target object in each first image sample;
dividing the first image sample into a training set and a verification set according to a preset proportion, wherein the number of the image samples in the training set is greater than that of the image samples in the verification set;
inputting the image samples in the training set into the target recognition model for training, verifying the target recognition model by using the verification set every other preset period, and verifying the accuracy of the target recognition model by using each first image sample in the verification set and the corresponding first target object; and
and when the verification accuracy is greater than a preset threshold value, finishing the training to obtain the trained target recognition model.
3. The method as claimed in claim 2, wherein the similarity algorithm comprises:
wherein x isiImage features, y, representing the first region imageiRepresenting image features of the second region image, d (x, y) representing xiAnd yiA similarity value between the first region image and the second region image.
4. The method for identifying the target identity based on the picture recognition as claimed in claim 1, wherein the integrated similarity value between the first target object and each second target object is calculated according to a preset calculation rule, and the calculation rule comprises a first calculation rule and a second calculation rule:
when the second image comprises a first area image with attribute information as a preset attribute, calculating a comprehensive similarity value of the first area image by adopting the first calculation rule, and taking the identity information of a second target corresponding to the comprehensive similarity value as the identity information of the first target;
and when the second image does not contain the first preset information, calculating a comprehensive similarity value of second preset information by adopting the second calculation rule, and taking the identity information of a second target corresponding to the comprehensive similarity value as the identity information of the first target.
5. The method for identifying an object based on image recognition according to claim 4, wherein the first calculation rule is:
wherein d represents a combined similarity value of the first and second objects, d1A similarity value between a first region image and a second region image representing that the attribute information is a preset attribute, wherein lambda represents a weight value of the first region image, and lambda represents a weight value of the first region image>0。
6. The method for identifying an object based on image recognition according to claim 4, wherein the second calculation rule is:
wherein d represents a combined similarity value of the first and second objects, diRepresenting the similarity value between the first region image and the second region image, m representing the number of the first region images, m being less than or equal to 3, i representing the order of the first region imagesIn the notation, λ represents a weight value of the first region image, and λ>0。
7. A server, comprising a memory and a processor, the memory having a target anomaly identification program stored thereon, the target anomaly identification program when executed by the processor implementing the steps of:
a receiving step: acquiring a first image of a first target object containing identity information to be recognized, inputting the first image into a pre-trained target recognition model, and outputting a second image;
an identification step: inputting the second image into a pre-trained feature extraction model, outputting a first image feature, inputting the first image feature into a pre-trained region identification model, identifying a first region image contained in the second image, and assigning a unique weight value to each type of the first region image;
a calculation step: inquiring a plurality of second area images matched with the attribute information of the first area image from a database of second area images of a second target object in which the identification information is stored in advance, and respectively calculating the similarity value between the first area image and each matched second area image; and
a confirmation step: and calculating the comprehensive similarity value of the first target object and each second target object according to the similarity value and the weight value of each first region image, and selecting the identity information of the second target object corresponding to the image with the largest comprehensive similarity value as the identity information of the first target object.
8. The server of claim 7, wherein the target recognition model is trained as follows:
acquiring first image samples, and marking the area of the first target object in each first image sample;
dividing the first image sample into a training set and a verification set according to a preset proportion, wherein the number of the image samples in the training set is greater than that of the image samples in the verification set;
inputting the image samples in the training set into the target recognition model for training, verifying the target recognition model by using the verification set every other preset period, and verifying the accuracy of the target recognition model by using each first image sample in the verification set and the corresponding first target object; and
and when the verification accuracy is greater than a preset threshold value, finishing the training to obtain the trained target recognition model.
9. The server of claim 8, wherein the similarity value algorithm comprises:
wherein x isiImage features, y, representing the first region imageiRepresenting image features of the second region image, d (x, y) representing xiAnd yiA similarity value between the first region image and the second region image.
10. A computer-readable storage medium having stored thereon a target anomaly identification program executable by one or more processors to implement the steps of the picture recognition based target identity recognition method according to any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010316554.8A CN111401326B (en) | 2020-04-21 | 2020-04-21 | Target identity recognition method based on picture recognition, server and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010316554.8A CN111401326B (en) | 2020-04-21 | 2020-04-21 | Target identity recognition method based on picture recognition, server and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111401326A true CN111401326A (en) | 2020-07-10 |
| CN111401326B CN111401326B (en) | 2023-04-18 |
Family
ID=71433334
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010316554.8A Active CN111401326B (en) | 2020-04-21 | 2020-04-21 | Target identity recognition method based on picture recognition, server and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111401326B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112488062A (en) * | 2020-12-18 | 2021-03-12 | 浙江大华技术股份有限公司 | Image identification method, device, equipment and medium |
| CN112507910A (en) * | 2020-12-15 | 2021-03-16 | 平安银行股份有限公司 | Image recognition method and system based on pixel deformation, electronic device and storage medium |
| CN112668479A (en) * | 2020-12-29 | 2021-04-16 | 广州耐奇电气科技有限公司 | Safety monitoring method and system for intelligent power distribution room, electronic equipment and medium |
| CN113516046A (en) * | 2021-05-18 | 2021-10-19 | 平安国际智慧城市科技股份有限公司 | Method, device, equipment and storage medium for monitoring biological diversity in area |
| CN117058715A (en) * | 2023-08-16 | 2023-11-14 | 北京观海科技发展有限责任公司 | Method for identifying identity information of moving object, method for generating moving object, equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150310601A1 (en) * | 2014-03-07 | 2015-10-29 | Digimarc Corporation | Methods and arrangements for identifying objects |
| CN108596277A (en) * | 2018-05-10 | 2018-09-28 | 腾讯科技（深圳）有限公司 | A kind of testing vehicle register identification method, apparatus and storage medium |
| CN109145927A (en) * | 2017-06-16 | 2019-01-04 | 杭州海康威视数字技术股份有限公司 | The target identification method and device of a kind of pair of strain image |
| WO2019132587A1 (en) * | 2017-12-29 | 2019-07-04 | (주)제이엘케이인스펙션 | Image analysis device and method |
| CN110516677A (en) * | 2019-08-23 | 2019-11-29 | 上海云绅智能科技有限公司 | A kind of neural network recognization model, target identification method and system |
| CN110619336A (en) * | 2019-08-26 | 2019-12-27 | 南京理工大学 | Goods identification algorithm based on image processing |
-
2020
- 2020-04-21 CN CN202010316554.8A patent/CN111401326B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150310601A1 (en) * | 2014-03-07 | 2015-10-29 | Digimarc Corporation | Methods and arrangements for identifying objects |
| CN109145927A (en) * | 2017-06-16 | 2019-01-04 | 杭州海康威视数字技术股份有限公司 | The target identification method and device of a kind of pair of strain image |
| WO2019132587A1 (en) * | 2017-12-29 | 2019-07-04 | (주)제이엘케이인스펙션 | Image analysis device and method |
| CN108596277A (en) * | 2018-05-10 | 2018-09-28 | 腾讯科技（深圳）有限公司 | A kind of testing vehicle register identification method, apparatus and storage medium |
| CN110516677A (en) * | 2019-08-23 | 2019-11-29 | 上海云绅智能科技有限公司 | A kind of neural network recognization model, target identification method and system |
| CN110619336A (en) * | 2019-08-26 | 2019-12-27 | 南京理工大学 | Goods identification algorithm based on image processing |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112507910A (en) * | 2020-12-15 | 2021-03-16 | 平安银行股份有限公司 | Image recognition method and system based on pixel deformation, electronic device and storage medium |
| CN112507910B (en) * | 2020-12-15 | 2023-01-17 | 平安银行股份有限公司 | Image recognition method and system based on pixel deformation, electronic device and storage medium |
| CN112488062A (en) * | 2020-12-18 | 2021-03-12 | 浙江大华技术股份有限公司 | Image identification method, device, equipment and medium |
| CN112488062B (en) * | 2020-12-18 | 2023-04-07 | 浙江大华技术股份有限公司 | Image identification method, device, equipment and medium |
| CN112668479A (en) * | 2020-12-29 | 2021-04-16 | 广州耐奇电气科技有限公司 | Safety monitoring method and system for intelligent power distribution room, electronic equipment and medium |
| CN113516046A (en) * | 2021-05-18 | 2021-10-19 | 平安国际智慧城市科技股份有限公司 | Method, device, equipment and storage medium for monitoring biological diversity in area |
| CN113516046B (en) * | 2021-05-18 | 2024-06-18 | 平安国际智慧城市科技股份有限公司 | Method, device, equipment and storage medium for monitoring biological diversity in area |
| CN117058715A (en) * | 2023-08-16 | 2023-11-14 | 北京观海科技发展有限责任公司 | Method for identifying identity information of moving object, method for generating moving object, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111401326B (en) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111401326B (en) | Target identity recognition method based on picture recognition, server and storage medium | |
| CN108764226B (en) | Image text recognition method, device, equipment and storage medium thereof | |
| CN111325104B (en) | Text recognition method, device and storage medium | |
| CN106446816B (en) | Face recognition method and device | |
| CN111160434B (en) | Training method and device for target detection model and computer readable storage medium | |
| CN110543815A (en) | Training method of face recognition model, face recognition method, device, equipment and storage medium | |
| CN110378235A (en) | A kind of fuzzy facial image recognition method, device and terminal device | |
| WO2016177259A1 (en) | Similar image recognition method and device | |
| CN110675940A (en) | Pathological image labeling method and device, computer equipment and storage medium | |
| US9633272B2 (en) | Real time object scanning using a mobile phone and cloud-based visual search engine | |
| US9767541B2 (en) | Techniques for automatically correcting groups of images | |
| CN110135889A (en) | Method, server and the storage medium of intelligent recommendation book list | |
| CN111476275A (en) | Image recognition-based target detection method, server and storage medium | |
| CN110008943B (en) | An image processing method and device, a computing device and a storage medium | |
| CN110826452B (en) | Palm print verification method and device, computer equipment and readable storage medium | |
| CN110705547B (en) | In-image character recognition method, device and computer-readable storage medium | |
| US20180349726A1 (en) | Logo image identification system | |
| WO2020172870A1 (en) | Method and apparatus for determining motion trajectory of target object | |
| JP2021184281A (en) | Verification system | |
| JP5278093B2 (en) | Article related information providing method, apparatus, program, and recording medium | |
| JP2018013887A (en) | Feature selection device, tag relevant area extraction device, method, and program | |
| CN111460816A (en) | Data retrieval method based on semantic features, electronic device and storage medium | |
| CN113591657B (en) | OCR layout recognition method and device, electronic equipment and medium | |
| CN110502658B (en) | Document image page number migration method, server and storage medium | |
| CN111737500B (en) | Electronic page retrieval method and device, learning equipment and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |