Identity recognition method, device and equipment, medium and program product

Abstract

The embodiment of the invention provides an identity recognition method and device, equipment, a medium and a program product. The method comprises the following steps: acquiring an action sequence of biological characteristics; performing feature extraction processing on each biological feature map in the plurality of biological feature maps to obtain a feature vector of each biological feature map; carrying out fusion processing on the feature vector of each biological feature map to generate a fused feature vector; and performing identity recognition on the biological features based on the fused feature vector to obtain a recognition result. According to the embodiment of the invention, richer feature information of the biological features can be extracted, and the recognition accuracy and reliability of the object to which the biological features belong are improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to an identity recognition method, an identity recognition device, a computer equipment, a computer-readable storage medium, and a computer program product. Background Art

[0002] With the progress of science and technology, the recognition of human biometrics is applied to various business scenarios to improve the convenience of business scenarios. For example, in the business scenario of transportation, biometrics such as palm prints or faces of users can be used for identity recognition and authentication to improve the convenience of the transportation process.

[0003] At present, when the feature recognition system collects the user's biometric features, there are often quality problems with the collected biometric features due to various reasons; such as low clarity, low resolution or incomplete feature collection of the biometric features. This makes it difficult to accurately identify the object to which the biometric features belong based on such low-quality images, thereby reducing the reliability of the recognition results. Summary of the invention

[0004] The embodiments of the present application provide an identity recognition method, apparatus, device, medium and program product, which can extract richer feature information of biometrics and improve the recognition accuracy and reliability of objects to which the biometrics belong.

[0005] On the one hand, an embodiment of the present application provides an identity recognition method, the method comprising:

[0006] An action sequence for acquiring biometric features; the action sequence includes a plurality of biometric feature graphs continuously collected for the biometric features; and the feature changes between the biometric features included in any two adjacent biometric feature graphs among the plurality of biometric feature graphs are continuous;

[0007] Performing feature extraction processing on each of the multiple biometric feature graphs to obtain a feature vector of each biometric feature graph; the feature vector of any biometric feature graph represents sub-feature information of the biometric feature from the semantic dimension of any biometric feature graph;

[0008] The feature vectors of each biometric feature graph are fused to generate a fused feature vector; the fused feature vector represents the comprehensive feature information of the biometric feature from the semantic dimension of the action sequence;

[0009] The biometric feature is identified based on the fused feature vector to obtain an identification result, which indicates the object to which the biometric feature belongs.

[0010] On the other hand, an embodiment of the present application provides an identity recognition device, the device comprising:

[0011] An acquisition unit for acquiring an action sequence of a biometric feature; the action sequence includes a plurality of biometric feature maps continuously acquired for the biometric feature; the feature changes between any two adjacent biometric feature maps among the plurality of biometric feature maps are continuous;

[0012] A processing unit for performing feature extraction processing on each biometric feature map among the plurality of biometric feature maps to obtain a feature vector of each biometric feature map; the feature vector of any biometric feature map represents sub-feature information of the biometric feature from the semantic dimension of any biometric feature map;

[0013] The processing unit is further configured to perform fusion processing on the feature vectors of each biometric feature map to generate a fused feature vector; the fused feature vector represents comprehensive feature information of the biometric feature from the semantic dimension of the action sequence;

[0014] The processing unit is further configured to perform identity recognition on the biometric feature based on the fused feature vector to obtain a recognition result, and the recognition result indicates the object to which the biometric feature belongs.

[0015] In one implementation, when the acquisition unit is configured to acquire an action sequence of a biometric feature, it is specifically configured to:

[0016] Continuously acquire the biometric feature of the target action being performed in the business scenario to obtain a video stream of the biometric feature; the video stream includes a plurality of initial biometric feature maps continuously acquired, and each initial biometric feature map includes all or part of the features of the biometric feature;

[0017] Preprocess the plurality of initial biometric feature maps included in the video stream to obtain an action sequence of the biometric feature;

[0018] Wherein, the biometric feature map is a palm map, the biometric feature included in the biometric feature map is a palmprint feature, and the business scenario is a long-distance palm-sweeping scenario.

[0019] In one implementation, the video stream includes N initial biometric feature maps, where N is an integer greater than 1; when the processing unit is configured to preprocess the plurality of initial biometric feature maps included in the video stream to obtain an action sequence of the biometric feature, it is specifically configured to:

[0020] Judge whether the feature change between the biometric feature included in the (i + 1)-th initial biometric feature map in the video stream and the biometric feature included in the i-th initial biometric feature map in the video stream is continuous; 1 ≤ i < N, and i is an integer;

[0021] If it is continuous, add the (i + 1)-th initial biometric feature map to the action sequence of the biometric feature, and judge whether the (i + 1)-th initial biometric feature map is the last one among the N initial biometric feature maps included in the video stream;

[0022] If the i+1th initial biometric feature map is not the last one of the N initial biometric feature maps contained in the video stream, then continue to determine whether the feature changes between the biometric features contained in the i+2th initial biometric feature map and the biometric features contained in the i+1th initial biometric feature map are continuous, until a candidate biometric feature map is detected in the video stream;

[0023] The biometric feature map includes: the last one of the N initial biometric feature maps contained in the video stream, or an initial biometric feature whose feature change between the contained biometric feature map and the biometric feature contained in the adjacent previous biometric feature map is not continuous.

[0024] In one implementation, the processing unit is further configured to:

[0025] If there is no continuity, then determine the action sequence of the biometric feature; the action sequence includes i initial biometric feature images in the video stream, and the i initial biometric feature images are: the initial biometric feature image before the i+1th initial biometric feature image in the video stream.

[0026] In one implementation, if a deep learning model is used to determine whether the feature changes between the biometric features included in the i+1th initial biometric feature graph and the biometric features included in the i-th initial biometric feature graph are continuous, the processing unit is further used to:

[0027] Add the i+1th initial biometric feature map to the interference data set;

[0028] Among them, the interference data set includes: an initial biometric feature map in which the feature changes between the biometric features included in the video stream and the previous adjacent initial biometric feature map are not continuous; the interference data set serves as interference data for the deep learning model during the model training process.

[0029] In one implementation, the processing unit is further configured to:

[0030] Performing image quality detection on each biometric feature image in the action sequence to obtain a quality detection result of each biometric feature image;

[0031] Filter out the biometric graphs whose quality inspection results do not meet the quality requirements from the action sequence;

[0032] Performing image enhancement processing on the biometric image whose quality inspection result does not meet the quality requirements to obtain an image-enhanced biometric image;

[0033] The biometric action sequence is updated by using the biometric image after image enhancement to obtain an updated action sequence; the updated action sequence includes: the biometric image after image enhancement and the biometric image whose quality detection result meets the quality requirements.

[0034] In one implementation, a biometric image whose quality inspection result does not meet the quality requirements is represented as a target biometric image; and a processing unit is used to perform image enhancement processing on the biometric image whose quality inspection result does not meet the quality requirements, and when obtaining the biometric image after image enhancement, is specifically used to:

[0035] Obtain the image processing requirements of the target biometric image and select an image enhancement algorithm that matches the image processing requirements;

[0036] Using an image processing algorithm that matches the image processing requirements, the target biological characteristic image is subjected to image enhancement processing to obtain the target biological characteristic image after image enhancement processing;

[0037] Among them, if the image processing requirement is to enhance the high-frequency information in the target biological feature map, the image enhancement algorithms that match the image processing requirements include: interpolation algorithms and learning algorithms under super-resolution algorithms.

[0038] In one implementation, the processing unit is used to perform identity recognition on the biometric feature based on the fused feature vector, and when obtaining the recognition result, is specifically used to:

[0039] Acquire a biometric feature data set, wherein the biometric feature data set includes candidate feature information of biometric features of different objects;

[0040] Using a feature comparison algorithm, the comprehensive feature information indicated by the fused feature vector is compared with each candidate feature information in the biometric feature data set to obtain a feature matching result corresponding to each candidate feature information;

[0041] Generate a recognition result based on the feature matching result corresponding to each candidate feature information;

[0042] The recognition result indicates that there is candidate feature information in the biometric data set whose feature matching result satisfies the matching condition, or the recognition result indicates that there is no candidate feature information in the biometric data set whose feature matching result satisfies the matching condition.

[0043] In one implementation, the feature comparison algorithm includes a similarity algorithm and a distance comparison algorithm; the processing unit is used to use the feature comparison algorithm to compare the comprehensive feature information indicated by the fused feature vector with each candidate feature information in the biometric feature data set, and obtain the feature matching result corresponding to each candidate feature information, specifically for:

[0044] Using a similarity algorithm, respectively, similarity calculations are performed on the comprehensive feature information indicated by the fused feature vector and each candidate feature information in the biometric feature data set to obtain a similarity comparison result corresponding to each candidate feature information; and,

[0045] By using a distance comparison algorithm, a distance index is calculated for each of the first distances between the biometric feature corresponding to the fused feature vector and the collection device when it is collected, and the second distance corresponding to each candidate feature information in the biometric feature data set, to obtain a distance index result corresponding to each candidate feature information; the distance index result is used to indicate the distance difference between the first distance and the second distance; the second distance corresponding to the candidate feature information refers to the distance between the biometric feature corresponding to the candidate feature information and the collection device when it is collected;

[0046] Based on the similarity comparison result and the distance index result corresponding to each candidate feature information, a feature matching result corresponding to each candidate feature information is generated.

[0047] In one implementation, when the feature matching result is a weighted value of a similarity comparison result and a distance index result, the feature matching result satisfies the matching condition including: the weighted value is greater than a weighted threshold;

[0048] When the feature matching result includes a similarity comparison result and a distance index result, the feature matching result satisfies the matching condition including: the similarity comparison result is greater than a similarity threshold, and the distance index result is greater than a distance difference threshold.

[0049] On the other hand, an embodiment of the present application provides a computer device, the computer device comprising:

[0050] A processor for loading and executing computer programs;

[0051] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned identity recognition method is implemented.

[0052] On the other hand, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program, and the computer program is suitable for being loaded by a processor and executing the above-mentioned identity recognition method.

[0053] On the other hand, an embodiment of the present application provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned identity recognition method is implemented.

[0054] In the embodiment of the present application, it is supported to continuously collect the biometric features of the user to obtain the action sequence of the biometric features; wherein the action sequence includes multiple biometric feature graphs obtained by continuous collection of the biometric features, and the feature changes between the biometric features contained in any two adjacent biometric feature graphs in the multiple biometric feature graphs are continuous. In this way, feature extraction processing can be performed on multiple biometric feature graphs in the action sequence to obtain a feature vector of each biometric feature graph; the feature vector of any biometric feature graph expresses the sub-feature information of the biometric feature from the semantic dimension of the biometric feature. For example, if the biometric feature is a palm, and the palm in the biometric feature graph is tilted, then the feature vector in the biometric feature graph represents the palm print characteristics when the palm is tilted. Furthermore, the feature vectors of each biometric feature graph are fused to generate a fused feature vector, which represents the comprehensive feature information of the biometric feature from the semantic dimension of the action sequence. Finally, identity recognition is performed based on the fused feature information of the biometric feature to obtain an identification result indicating the object to which the biometric feature belongs. It can be seen from the above scheme that the embodiment of the present application provides a new biometric collection and identification scheme, which can continuously collect multiple biometric graphs of the user when performing the target action (such as palm swiping action / palm swiping behavior), and analyze the comprehensive and rich vector representation of the biometrics based on multiple biometric graphs for identity identification; compared with identity identification based on the feature vector of a single biometric graph, it effectively improves the accuracy and reliability of the biometrics used for identity identification, thereby improving the accuracy of identity identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 is a structural schematic diagram of a feature recognition system provided by an exemplary embodiment of the present application;

[0057] Figure 2 It is a schematic diagram of the architecture of a business system provided by an exemplary embodiment of the present application;

[0058] Figure 3 It is a flowchart of an identity recognition method provided by an exemplary embodiment of the present application;

[0059] Figure 4 It is a schematic diagram of a video stream for collecting biometric features provided by an exemplary embodiment of the present application;

[0060] Figure 5 is a schematic diagram of filtering out a biometric feature graph constituting an action sequence from a video stream provided by an exemplary embodiment of the present application;

[0061] Figure 6 It is a flowchart of feature extraction and feature fusion provided by an exemplary embodiment of the present application;

[0062] Figure 7 is a flowchart of another identity recognition method provided by an exemplary embodiment of the present application;

[0063] Figure 8 It is a schematic diagram of a process for optimizing image quality for a biometric feature image provided by an exemplary embodiment of the present application;

[0064] Fig. 9 is a structural schematic diagram of an identity recognition device provided by an exemplary embodiment of the present application;

[0065] Fig.10 It is a structural diagram of a computer device provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0067] The embodiments of the present application relate to biometric identification technology; biometric identification technology refers to technology that uses biological characteristics inherent in organisms to perform identity recognition or authentication through computer equipment. Among them, the biological characteristics inherent in organisms may include but are not limited to: palm prints, fingerprints, faces, irises and retinas, etc.; the biological characteristics of each organism are unique, so the identity recognition of the organism can be realized based on the biological characteristics. For example, palm print recognition refers to the recognition technology that uses the palm print characteristics of the palm (or sole) of the organism to realize identity recognition; taking the palm of the organism as an example, the palm print of the palm of the organism refers to the palm characteristics from the end of the finger to the wrist. Among them, there are many features in the palm characteristics of the organism that can be used for identity recognition, such as the palm print main line, palm print wrinkles, palm print texture or palm print bifurcation points displayed in the palm can be used to realize palm print recognition. For example, iris recognition refers to the recognition technology based on the iris in the eye for identity recognition; mainly considering the characteristics that the iris remains unchanged throughout the life course, which determines the uniqueness of the iris characteristics.

[0068] It is worth noting that the embodiments of the present application do not limit whether the organism to which the biometric belongs is an animal or a human. In practical applications, the type of organism to which the biometric belongs varies according to the different application fields of the biometric recognition technology. For example, in the field of wild animal recognition, the biometric belongs to an animal. For the sake of ease of explanation, the embodiments of the present application take the biometric as a palm print and the organism to which the biometric belongs is a human as an example to introduce the subsequent content, which is specially explained here.

[0069] Furthermore, with the development of biometric recognition technology, non-invasive biometric recognition technology has also emerged; the so-called non-invasiveness of biometric recognition means that for a living being, the entire process of collecting and identifying the biological characteristics of the living being is non-contact, even imperceptible or silent. This non-contact biometric collection and recognition method has low requirements on the performance of the collection and recognition equipment, and does not require direct contact with the living being. It can achieve rapid or multi-biological collection and recognition in a short time, and has high recognition efficiency. However, it has been found in practice that the above-described non-contact biometric collection and recognition method may result in poor quality of the collected biometrics due to various reasons, thereby greatly reducing the accuracy and reliability of identity recognition. For example, in the palm print recognition process, if the user's palm print is collected at a long distance, it is likely that the resolution of the collected palm print image is low due to the long distance, and the clarity of the palm print contained in the palm print image is low. If the identity recognition is based on the low-resolution and low-definition palm print image, it is bound to cause poor reliability of the identity recognition result.

[0070] In order to improve the reliability of biometric features and ensure the accuracy of identity recognition, the embodiment of the present application provides an identity recognition solution; the solution combines technologies such as continuous video stream acquisition and feature fusion to improve the accuracy and reliability of biometric feature recognition. Specifically, the general process of the solution may include:

[0071] First, obtain the action sequence of biometrics; the action sequence is collected when the user performs the target action in front of the acquisition device, such as when the user performs the palm swiping action in front of the palm print acquisition device, and the action sequence includes multiple biometric graphs obtained by continuous collection of biometrics, and the feature changes between the biometrics contained in any two adjacent biometric graphs in the multiple biometric graphs are continuous, that is, the biometric changes presented by the two adjacent biometric graphs are orderly (according to the execution order of the target action). Then, perform feature extraction processing on each of the multiple biometric graphs included in the action sequence to obtain a feature vector of each biometric graph; the feature vector of any biometric graph represents the sub-feature information of the biometric from the semantic dimension of the biometric graph. Secondly, the feature vectors of each biometric graph are fused to generate a fused feature vector; compared with the feature vector of a single biometric graph, the fused feature vector represents the comprehensive feature information of the biometric from the semantic dimension of the action sequence. Finally, based on the comprehensive fused feature information, the identity recognition of the biometrics can obtain a highly accurate recognition result.

[0072] It can be seen that the embodiment of the present application can utilize the feature complementarity (or feature integrity, continuity of feature changes, etc.) between multiple biometric feature maps collected continuously, superimpose to obtain a richer and more comprehensive fusion feature vector of the biometric feature, and fuse the feature vector for identity recognition. Compared with identity recognition based on the feature vector of a single biometric feature map, the accuracy and reliability of the biometric features used for identity recognition are effectively improved, thereby improving the accuracy of identity recognition. In addition, the recognition scheme provided by the embodiment of the present application is not limited by the distance between the user and the collection device. Even if the user's biometric features are collected at a long distance, comprehensive feature information of the biometric features can be obtained, effectively filling the technical gap in the field of long-distance feature collection, and providing effective support for the application of biometric recognition technology. Provide reliable support for biometric recognition applications in fields such as security authentication, and bring a higher level of security and convenience to society.

[0073] Based on the above brief introduction to the identity recognition solution provided in the embodiment of the present application, the following points need to be explained:

[0074] (1) The identity recognition solution provided in the embodiments of the present application can be deployed in a feature recognition system (or platform) with biometric feature collection and recognition functions.

[0075] Among them, the feature recognition system may refer to an application that supports the collection and recognition of biometric features. An application may refer to a computer program for completing one or more specific tasks; by classifying the application according to different dimensions (such as the operation mode and function of the application), the types of the same application under different dimensions can be obtained. For example: according to the operation mode of the application, the application may include but is not limited to: a client installed in the terminal, a small program (as a subprogram of the client) that can be used without downloading and installing, a web (World Wide Web) application opened through a browser, etc. For another example: according to the functional type of the application, the application may include but is not limited to: IM (Instant Messaging) application, content interaction application, etc. Among them, an instant messaging application refers to an application for instant communication of messages and social interaction based on the Internet. The instant messaging application may include but is not limited to: a social application with communication functions, a map application with social interaction functions, a game application, etc. Content interaction application refers to an application that can realize content interaction, such as online banking, sharing platforms, personal space, news and other applications.

[0076] Furthermore, the feature recognition system can also be a plug-in (or function) included in the above-mentioned application that supports biometric collection and identification. For example, if the application is the instant messaging application mentioned above, then the feature recognition system can be a biometric plug-in included in the instant messaging application; if the user needs to collect and identify biometrics during a conversation using the instant messaging application, the biometric collection and identification plug-in can be directly called from the instant messaging application to implement the biometric collection and identification function; in this way, users (such as any object using the instant messaging application) can also perform functions such as biometric collection and identification while using the instant messaging application for socializing, without the need for application jumps (such as jumping from the instant messaging application to an independent biometric collection and identification application).

[0077] (2) The identification scheme provided in the embodiments of the present application can be applied to any business scenario that requires biometric collection and identification. The fields involved in the business scenarios may include but are not limited to: security, transportation, education, payment, and personal identity authentication scenarios, etc. Taking the biometric map as a palm map, the biometric features contained in the biometric map as palm print features (or simply palm prints), and the business scenario as a long-distance palm swiping scenario as an example, exemplary business scenarios in different fields may include but are not limited to:

[0078] Optionally, business scenarios in the security field may include security access control scenarios. In the security access control scenario, users who are waiting to enter the company can perform palm swiping while standing at a fixed location at a distance or walking. At this time, the acquisition device (i.e., a computer device deployed with the identity recognition solution (or the feature recognition system deployed by the identity recognition solution) provided in the embodiment of the present application) can continuously collect palm swiping behaviors to obtain a video stream; and, feature extraction and feature fusion are performed on the continuous multiple biometric feature maps included in the video stream, so that the user will be allowed to enter the company only when the acquisition device determines that the collected palm print belongs to an employee within the company.

[0079] Optionally, business scenarios in the transportation field may include ticket checking scenarios. In the ticket checking scenario, users who are waiting to enter the station can perform palm scanning at a fixed location or while walking. At this time, the collection device will allow the user to enter the station only when it detects that the user whose palm print is collected has purchased a ticket and the ticket has not been used.

[0080] Optionally, business scenarios in the education field may include clocking-in scenarios. In the clocking-in scenario, the teacher can perform palm scanning at a fixed location or while walking before each class. At this time, the collection device can associate the collected palm prints with the teacher and record the teacher's class behavior in a timely manner.

[0081] Optionally, the business scenarios in the payment field include offline transfer scenarios. In offline transfer scenarios, consumers can perform palm scanning while standing at a fixed location at a distance or walking. In this way, after the collection device collects the consumer's palm print, it can find the consumer's account from the database based on the palm print, and reduce the amount of resources corresponding to the goods purchased by the consumer from the resource pool corresponding to the consumer's account, and increase the amount of resources corresponding to the goods purchased by the consumer in the resource pool of the store (i.e., the user to be paid) to achieve secure payment.

[0082] It should be understood that, considering that the process and principle of implementing the collection and identification of biometric features in various biometric feature identification scenarios (or business scenarios) using the identity recognition scheme provided in the embodiment of the present application are similar, only the types of biometric features involved in different identification scenarios may be different. Therefore, the above is only an explanation of the general query principles in various business scenarios, and the relevant descriptions of each business scenario will not limit the embodiments of the present application, which is specifically explained here. For example, business scenarios in the security field can also include detection scenarios; for example, in the detection scenario, the video stream about a certain user collected by the acquisition device is often low-resolution. If feature recognition is performed based only on a single biometric feature map in the video stream, the accuracy and reliability of the recognition are undoubtedly very low; therefore, the recognition scheme provided in the embodiment of the present application can be used to fuse the feature vectors of multiple biometric feature maps to obtain a fused feature vector, so that it is easier to identify the user's identity based on a comprehensive and comprehensive fused feature vector.

[0083] It can be seen that compared with the traditional method that requires users to constantly adjust the distance between themselves and the collection device, and adjust the collection angle after the position is fixed so that the collection device can collect a biometric image with better quality; the identity recognition solution provided in the embodiment of the present application does not require the user to remain still during the process of biometric collection and recognition, nor does it limit the distance between the user and the collection device. On the basis of ensuring the accuracy and reliability of biometric recognition, it effectively improves the speed and efficiency of biometric collection and recognition.

[0084] (3) The structure of the feature recognition system involved in the embodiment of the present application may include but is not limited to four parts. Figure 1 As shown, they are: biometric feature acquisition subsystem, feature processing subsystem, feature extraction and fusion subsystem and biometric feature recognition subsystem; among which:

[0085] ① The biometric acquisition subsystem has the function of acquiring multiple feature maps through video streams, which refers to the technology of continuously acquiring multiple biometric feature maps through video streams; more palmprint information is acquired through continuous acquisition to improve acquisition efficiency and accuracy. The biometric acquisition subsystem includes: a continuous feature map capture module based on video streams and a map storage module. The continuous feature map capture module is designed to use a special action camera to continuously acquire a video stream of the user's palm swiping behavior, that is, the technology of continuously acquiring multiple palmprint images through video streams is used to acquire more palmprint information through continuous acquisition to improve acquisition efficiency and accuracy. The map storage module is mainly used to store continuous biometric feature maps in video streams.

[0086] ② The feature processing subsystem includes: an optimization module and a graph storage module. The optimization module can optimize low-resolution images in a plurality of consecutive biometric graphs, such as using a super-resolution algorithm. The graph storage module is used to store optimized biometric graphs. Among them, super-resolution technology is a technology that improves image resolution through image processing algorithms, which can restore clearer and more detailed high-resolution images from low-resolution images, thereby improving image quality and recognition accuracy.

[0087] ③The feature extraction and fusion subsystem includes: feature extraction module, feature storage module and feature fusion module. The feature extraction module is used to perform feature extraction processing on each biometric feature map. The feature storage module is used to store the feature vector of each biometric feature map after feature extraction. The feature fusion module has the function of feature fusion, which means fusing the feature vectors of multiple biometric feature maps to form a more comprehensive and accurate feature representation. By comprehensively utilizing the features of multiple biometric feature maps, the performance and robustness of the feature recognition system can be improved, and the ability to identify and distinguish individuals can be enhanced; the feature fusion module is used to fuse the feature vectors of multiple biometric feature maps to obtain more complete and rich comprehensive features.

[0088] ④ The biometric identification subsystem has biometric identification technology; for example, palmprint recognition technology refers to the technology of using the shape and texture of palmprints to identify individuals. By analyzing and comparing the features of palmprint images, it can confirm and identify personal identities. It has uniqueness, stability and high reliability. The biometric identification subsystem includes a biometric identification module, which is mainly used to perform palmprint recognition on the fused feature vector after fusion to obtain the recognition result.

[0089] In a specific implementation, when there is a need to collect and identify the biometric features of an object, the biometric features of the user (i.e., the object) can be continuously collected through the biometric feature collection subsystem (specifically, the continuous feature map capture module deployed in the subsystem) to capture a video stream; and the biometric feature collection subsystem (specifically, the map storage module deployed in the subsystem) can cache multiple biometric maps with continuity in the video stream. Then, the feature processing subsystem (specifically, the optimization module deployed in the subsystem) can optimize the multiple biometric maps collected, such as using a super-resolution algorithm for optimization; and the optimized biometric maps can be stored through the feature processing subsystem (specifically, the map storage module deployed in the subsystem). Then, the feature extraction and fusion subsystem (specifically, the feature extraction module deployed in the subsystem) can perform feature extraction processing on each optimized biometric feature map to obtain a feature vector of each biometric feature map; and the feature extraction and fusion subsystem (specifically, the feature storage module deployed in the subsystem) caches the feature vector of each biometric feature map; in addition, the feature extraction and fusion subsystem (specifically, the feature fusion module deployed in the subsystem) performs fusion processing on the feature vector of each biometric feature map to generate a fused feature vector. Finally, the palmprint recognition subsystem (specifically, the feature recognition module deployed in the subsystem) feature recognition module can realize the recognition processing of the biometric feature based on the fused feature vector to obtain a recognition result, which indicates the object to which the biometric feature belongs.

[0090] It should be noted that the modules included in the feature recognition system involved in the embodiment of the present application are not limited to the modules given above. For example, a continuity judgment module is also deployed in the feature acquisition subsystem, which has the function of initializing the video stream. The initialization processing here may include but is not limited to: judging the continuity between adjacent biometric feature graphs included in the video stream (such as judging whether the current biometric feature graph is the next biometric feature graph of the previous biometric feature graph in the process of the object performing a coherent target action) to ensure the continuity between multiple biometric feature graphs, so that the continuity can be used to obtain comprehensive and integrated feature information of the biometrics; and archiving the biometric feature graph after the continuity judgment according to the sequence number of the action sequence. Of course, the function of the continuity judgment module can also be incorporated into the biometric feature acquisition subsystem given above, or incorporated into the feature processing subsystem; the embodiment of the present application does not limit this.

[0091] (4) The identity recognition solution provided in the embodiment of the present application can be executed by a computer device, and the embodiment of the present application does not limit the number and type of computer devices. Figure 2 is a schematic diagram of the architecture of a business system provided by an exemplary embodiment of the present application; Figure 2As shown, the business system includes computer device 201 and computer device 202. The embodiment of the present application does not limit the number and naming of computer device 201 and computer device 202. Among them:

[0092] ① The computer device 201 may refer to a device having a biometric feature collection function. Specifically, the computer device 201 is equipped with Figure 1 The biometric feature collection subsystem shown. The computer device may be a terminal device that directly interacts with the user; the terminal device may include but is not limited to: a smart phone (such as a smart phone with an Android system, or a smart phone with an Internetworking Operating System (IOS)), a tablet computer, a portable personal computer, a mobile Internet device (Mobile Internet Devices, MID for short), a vehicle-mounted device, a head-mounted device, and other devices. The embodiment of the present application does not limit the type of the terminal device, which is explained here.

[0093] ② Computer device 202 is a server corresponding to computer device 201, and is used to interact with computer device 201 to provide computing and application service support for the video stream sent by computer device 201. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. Furthermore, a feature processing subsystem, a feature extraction and fusion subsystem are deployed in the server; in this way, after receiving the video stream collected by computer device 201, the server can use the feature processing subsystem to pre-process the video stream, so that the pre-processed video stream is more suitable for implementing subsequent operations, thereby improving the quality of the pre-processed biometric feature map. The feature extraction and fusion subsystem in the server is used to perform feature extraction processing on the preprocessed biometric feature image to obtain a feature vector of the biometric feature image, and to fuse the feature vectors of multiple biometric feature images to generate a fused feature vector that can comprehensively characterize the characteristics of the biometric feature.

[0094] Furthermore, the server also includes a biometric identification subsystem and a database 203 (or a biometric data set). The database 203 can be used to store candidate feature information of biometrics of different objects (i.e., users); in this way, after obtaining the fused feature information of the biometrics, the server can use the biometric identification subsystem to identify the biometrics, specifically, using the fused feature vector of the biometrics to find candidate feature information that matches (or is the same as) the comprehensive feature information indicated by the fused feature information of the biometrics from the database 203, so that the object to which the candidate feature information belongs is used as the object to which the biometrics to be identified belongs, so as to achieve the purpose of identifying the object. In addition, if the computer device 201 needs to receive feedback information from the computer device 202 before releasing the user (such as a security access control scenario), then after obtaining the recognition result, the computer device 202 will also return feedback information to the computer device 201 according to the recognition result, and the feedback information indicates that the computer device 201 responds to the user's palm-swiping behavior (or other behavior) (such as opening the door or prompting payment completion, etc.).

[0095] It should be understood that the present embodiment does not limit which subsystems of the feature recognition system are specifically deployed in the computer device 101 and the computer device 102. Figure 2 The example in which the identity recognition scheme can be jointly executed by the computer device 201 and the computer device 202 is used for explanation; however, in actual application, when the computer device 201 has the function of biometric recognition, Figure 2 The system shown may only include a computer device 201. In this case, the identity recognition solution provided in the embodiment of the present application may be executed only by the computer device 201. Then, all subsystems included in the feature recognition system are deployed in the computer device 201.

[0096] (5) The collection and processing of relevant data in the embodiments of this application should be strictly in accordance with the requirements of the laws and regulations of relevant countries. The acquisition of personal information must be informed or agreed to by the individual subject (or have a legal basis for information acquisition), and the subsequent use and processing of data shall be carried out within the scope of authorization of laws and regulations and the subject of personal information. For example, when the embodiments of this application are applied to specific products or technologies, such as collecting the user's biometric features, the user's permission or consent is required, and the collection, use and processing of relevant data (such as the collection and release of the bullet comments posted by the subject, etc.) must comply with the relevant laws, regulations and standards of relevant countries and regions.

[0097] Based on the identity recognition scheme described above, the embodiment of the present application proposes a more detailed identity recognition method. The identity recognition method proposed in the embodiment of the present application will be described in detail below in conjunction with the accompanying drawings.

[0098] Figure 3 A schematic diagram of a flow chart of an identity recognition method provided by an exemplary embodiment of the present application is shown; Figure 3 The identity recognition method shown may be executed by a computer device, and the identity recognition method may include but is not limited to steps S301-S304:

[0099] S301: Acquire an action sequence of biometric features.

[0100] The action sequence of the biometrics includes: multiple biometric graphs obtained by continuous collection of the biometrics, and the feature changes between any two adjacent biometric graphs in the multiple biometric graphs included in the action sequence are continuous; that is, from the overall point of view, the changes in the biometrics presented by the action sequence are continuous. It is precisely relying on the continuity (or coherence) between the multiple biometric graphs included in the action sequence that a more comprehensive and reliable fusion feature vector of the biometrics is obtained based on the feature complementarity between the continuous multiple biometric graphs, so that the accuracy of identity recognition can be improved based on the fusion feature vector. Among them, taking the biometric as a palm print (or palm print feature) as an example, the continuity of the feature changes between the palm prints included in the multiple biometric graphs included in the action sequence can be simply understood as: according to the palm print collection rules when the user performs the palm swiping behavior, the collection device will first collect the fingertips, then collect the fingers, then collect part of the palm, until the complete palm is collected. It can be seen from this that when the user performs palm scanning, the changes in the palm prints presented by the multiple continuously collected biometric feature images should be a continuous process from part to whole.

[0101] In actual applications, if there is a need to perform biometric identification on a user in a business scenario, the action sequence of the user's biometrics can be obtained to implement subsequent feature extraction, feature fusion, and feature identification based on the action sequence to accurately identify the user's identity. The specific implementation process of obtaining the action sequence of the user's biometrics may include but is not limited to:

[0102] (1) When the acquisition device is turned on, the acquisition device continuously acquires the biometric features of the target action in the business scenario to obtain a video stream about the biometric features. At this time, the acquired video stream includes multiple initial biometric feature images acquired continuously, and each initial biometric feature image may include all or part of the features of the biometric features (such as a complete palm or a partial palm (such as only the finger part)). In other words, when the acquisition device is turned on, the process of the acquisition device acquiring the video stream about the biometric features is similar to the process of taking a video of the biometric features, and the acquired video stream reflects the continuous process of the user's biometric features being used to perform the target action.

[0103] Taking the palm print as an example, when the collection device is turned on, the continuous process of the user performing the target action (such as palm swiping action, or palm swiping behavior) is collected to obtain a video stream about the user's palm print. Figure 4 As shown, when the acquisition device is in the on state, as the user performs the palm swiping behavior, the user's finger first appears in the acquisition range of the acquisition device, then the acquisition device first acquires the initial biometric characteristic map of the fingertips of the fingers including the palm; as the user continues to perform the palm swiping behavior, the acquisition device acquires the initial biometric characteristic map of the fingers including the palm; continuing, as the user continues to perform the palm swiping behavior, the acquisition device acquires the initial biometric characteristic map including part of the palm; and so on, until the acquisition device meets the stop acquisition condition, N initial biometric characteristic maps acquired by the acquisition device are obtained, N is an integer greater than 3 (in other cases, N can be an integer greater than 1). Then, it is determined that the video stream acquired by the acquisition device for the user's palm print includes N initial biometric characteristic maps.

[0104] The embodiments of the present application do not limit the specific content of the above-mentioned conditions for the collection device to stop collection. Optionally, the collection device starts timing from the detection of the user's biometric characteristics, and after continuously collecting the biometric characteristics for a target time period, the collection device automatically changes from an on state to a off state, thereby stopping collection; if the target time period is 10 seconds, when the collection device detects that the user's biometric characteristics appear within the collection range, collection begins and continues for 10 seconds, and within these 10 seconds, even if the user's biometric characteristics leave the collection range, the collection device will continue to collect. Optionally, when the collection device detects that the user's biometric characteristics fall into the collection range, it begins to collect the biometric characteristics until the biometric characteristics leave the collection range, at which time the collection device stops collecting.

[0105] It can be seen that in the video stream obtained by the acquisition device continuously acquiring the biometric features of the target action being executed, the feature changes between adjacent different initial biometric maps in multiple initial biometric maps may be relatively large. This will cause the feature changes to be discontinuous and non-complementary, so that when the subsequent feature fusion is performed based on the features that are not discontinuous and non-complementary, the reliability of the fused feature vector obtained is low. For example, if the user quickly changes from spreading the palm to clenching the palm (such as making a fist), the feature change between the two adjacent initial biometric maps will be too large, resulting in poor subsequent feature fusion effects. For another example, when collecting the user's palm prints, if the user shakes his palm quickly, it will also cause the feature change between the two adjacent initial biometric maps to be too large, resulting in poor subsequent feature fusion effects.

[0106] (2) In order to separate multiple biometric feature graphs with strong continuity from the video stream (i.e., multiple initial biometric feature graphs screened from the video stream) to form an action sequence and improve the reliability of the biometric feature. After collecting the video stream of the biometric feature, the embodiment of the present application preprocesses the multiple initial biometric feature graphs included in the video stream to obtain a biometric action sequence; the feature changes between the multiple biometric feature graphs included in the action sequence have strong continuity and complementarity.

[0107] Among them, the specific implementation logic of preprocessing the video stream to obtain a more continuous action sequence may include but is not limited to: in a business scenario, after using a special action camera to continuously capture the target action performed by the user to obtain a video stream, the feature change requirements of the business scenario can be obtained, and the feature change requirements stipulate: the requirements that the feature changes between the current initial biometric feature map and the adjacent previous initial biometric feature map that can be added to the action sequence in the video stream must meet. Then, according to the order between the initial biometric feature maps included in the video stream, it is judged in turn whether the feature changes between the biometric features contained in the two adjacent initial biometric feature maps meet the feature change requirements of the business scenario. If they meet, the two adjacent initial biometric feature maps are added to the action sequence; if they do not meet, the initial biometric feature map located at the back position of the two adjacent initial biometric feature maps and the initial biometric feature map after the initial biometric feature map located at the back position in the video stream are not added to the action sequence.

[0108] For example, a video stream includes N initial biometric feature images, where N is an integer greater than 1. Assume that the feature change requirement of the business scenario can be: the feature changes of the biometric features contained in two adjacent initial biometric feature images in the video stream are continuous. Then, Figure 5As shown, after continuously collecting a video stream by using a dedicated action camera for the target action performed by the user, it is possible to determine whether the feature change between the biometric features included in the (i + 1)-th initial biometric map in the video stream and the biometric features included in the i-th initial biometric map in the video stream is continuous; 1 ≤ i < N, and i is an integer. If the feature change between the biometric features included in the (i + 1)-th initial biometric map in the video stream and the biometric features included in the i-th initial biometric map in the video stream is continuous, indicating that there is action coherence between the biometric features included in the i-th biometric map and the biometric features included in the (i + 1)-th biometric map, which can be used to achieve feature fusion and recognition, then add the (i + 1)-th initial biometric map to the action sequence of biometric features. And continue to determine whether there is an (i + 2)-th initial biometric map in the video stream (that is, determine whether the (i + 1)-th initial biometric map is the last one among the N initial biometric maps included in the video stream). If it exists (that is, the (i + 1)-th initial biometric map is not the last one among the N initial biometric maps included in the video stream), then continue to determine whether the biometric feature change between the biometric features included in the (i + 2)-th initial biometric map and the biometric features included in the (i + 1)-th initial biometric map is continuous until a candidate biometric map in the video stream is detected; the candidate biometric map includes: the last one among the N initial biometric maps included in the video stream, or, an initial biometric map whose feature change between the biometric features included in the biometric map and the biometric features included in the adjacent previous biometric map is not continuous.

[0109] Conversely, if the feature change between the biometric features included in the (i + 1)-th initial biometric map in the video stream and the biometric features included in the i-th initial biometric map in the video stream is not continuous (such as Figure 5 as shown, the palm included in the (i + 1)-th initial biometric map is in a clenched fist state, while the palm included in the i-th initial biometric map is in an open state, then the two are not continuous), indicating that the biometric features included in the (i + 1)-th biometric map have a relatively large action change compared to the biometric features included in the i-th biometric map, without coherence, and cannot be used to achieve feature fusion and recognition; then determine the action sequence of biometric features. At this time, the action sequence includes the i initial biometric maps in the video stream, and the i initial biometric maps are: the initial biometric maps in the video stream that are before the (i + 1)-th initial biometric map.

[0110] It should be understood that the purpose of preprocessing the video stream is to ensure that the feature changes between the multiple biometric graphs contained in the action sequence are continuous; therefore, the embodiment of the present application does not limit the specific algorithm used for preprocessing the video stream. For example, the embodiment of the present application supports the use of a deep learning model to determine whether the feature changes between the biometrics contained in two adjacent initial biometric graphs in the video stream are continuous. In other words, the trained deep learning model can learn and understand the action relationship between the biometrics contained in different initial biometric graphs, that is, to determine whether the action presented by the biometric contained in an initial biometric graph is the next action of the action presented by the biometric contained in the adjacent previous initial biometric graph. The embodiment of the present application does not limit the model structure and type of the deep learning model, such as the deep learning model may include but is not limited to: one or more of: Long Short-Term Memory (LSTM), Artificial Neural Network (ANN) and Multilayer Perceptron (MLP).

[0111] Further, such as Figure 2The computer device 202 in the system shown can deploy the deep learning model mentioned above; in this implementation, after the computer device 201 collects the video stream through a special action camera, the continuously collected video stream can be transmitted to the computer device 202, and the computer device 202 calls the trained deep learning model to make continuity judgments on the adjacent initial biometric feature maps included in the video stream. Among them, if the feature changes between the biometric features contained in two adjacent initial biometric feature maps are continuous, then the two adjacent initial biometric feature maps can be archived to facilitate adding to the action sequence of the biometric features, ensuring the continuity and integrity between the multiple biometric feature maps included in the action sequence. On the contrary, if the feature changes between the biometric features contained in the two adjacent initial biometric feature maps are not continuous (such as the feature changes between the biometric features contained in the i-th initial biometric feature map and the biometric features contained in the i+1-th initial biometric feature map are not continuous), that is, the action is interrupted, then the i+1-th initial biometric feature map can be appropriately processed to add the processed i+1-th initial biometric feature map to the interference data set. The interference data set includes: an initial biometric map whose biometric changes between the biometrics contained in the video stream and the previous adjacent initial biometric map are not continuous; the interference data set can be used as interference data for the deep learning model during the model training process. In the above method, the initial biometric map corresponding to the interrupt action is properly processed (such as labeling, regularization, and dimensionality reduction) and then cached into the interference data set, which can effectively utilize the collected initial biometric map, which is not only conducive to the interference training of the deep learning model, but also avoids the waste of the collected initial biometric map.

[0112] To summarize, through the video stream multi-feature map acquisition technology, the user's target action can be captured in real time and continuously, ensuring the continuity between the initial biometric maps contained in the acquired video stream; and the continuity of the feature changes between the biometrics contained in adjacent initial biometric maps in the video stream is judged to construct a biometric action sequence, ensuring the continuity, complementarity and integrity between the biometrics contained in multiple biometric maps contained in the action sequence, thereby obtaining richer feature information of the user and improving the reliability and authenticity of the biometrics.

[0113] S302: Perform feature extraction processing on each of the multiple biological feature graphs to obtain a feature vector of each biological feature graph.

[0114] S303: Fusing the feature vectors of each biological feature map to generate a fused feature vector.

[0115] In steps S302-S303, after obtaining the biometric action sequence based on the aforementioned step S301, the continuous feature graph feature extraction and fusion module can be used to extract and fuse the biometric graphs in the action sequence, aiming to use multiple continuous biometric graphs to extract and fuse biometrics, so as to further improve the accuracy and reliability of biometric recognition. Among them, the feature extraction and fusion of the biometric graphs in the action sequence can include two parts, namely: feature extraction for each biometric graph in the action sequence, and feature fusion of feature vectors for multiple biometric graphs. Among them:

[0116] ① Feature extraction for each biometric graph in the action sequence aims to remove redundant information and noise in the original data in the biometric graph, thereby reducing the dimension of the data, which is conducive to analyzing the key features of the biometric graph (such as color, lines and patterns) and improving the data processing efficiency. Furthermore, feature extraction can also map the biometric graph into a feature space, and use a vector form to represent the semantics of the biometric graph to extract a feature vector representing the semantic features of the biometric graph. The embodiment of the present application does not limit the type of algorithm for feature extraction for biometric graphs. For example, the embodiment of the present application supports training a deep learning model with a feature extraction function, which can extract features from continuous biometric graphs after training; wherein, the deep learning model with a feature extraction function may include but is not limited to: Recurrent Neural Network (RNN), Convolutional Neural Networks (CNN) and Transformer network, etc.

[0117] Recurrent neural network is a type of neural network with short-term memory. Neurons in recurrent neural network can not only receive information from other neurons, but also receive their own information, forming a network structure with loops. Convolutional neural network is a type of feedforward neural network that includes convolution calculations. It is designed based on the translation invariance of image tasks (the objects recognized by image have the same meaning in different positions). Transformer network is a neural network model based on attention mechanism. Transformer network abandons the convolution model structure and only achieves good performance through attention mechanism and feedforward neural network without using sequence alignment recurrent architecture. Among them, attention mechanism is a solution proposed by imitating human attention. Simply put, it is to imitate human attention to quickly filter out the information you want to pay attention to from a large amount of information. It is mainly used to solve the problem that it is difficult to obtain a reasonable vector representation when the input sequence of the time series model is long. The method is to retain the intermediate results of the time series model, learn it with a new model and associate it with the output, so as to achieve the purpose of information screening.

[0118] ② Feature fusion of feature vectors of multiple biometric maps refers to the process of fusing feature vectors of multiple biometric maps to form a more comprehensive and accurate fused feature vector. Feature fusion aims to fuse the complete and rich feature information extracted from each biometric map included in the action sequence to obtain a fused feature vector that can represent the important information of all biometric maps; the fused fused feature vector can contain more comprehensive and rich biometric information, improve the reliability and accuracy of biometrics, and improve the performance of biometric recognition. The method of feature fusion of feature vectors of multiple biometric maps in the action sequence may include but is not limited to: feature connection, feature weighting, etc. The so-called feature connection refers to the direct feature splicing of feature vectors of multiple biometric maps. If the dimensions of the feature vectors of two biometric maps are p and q respectively, the dimension of the feature vector after feature splicing is p+q. The so-called feature weighting refers to adding weights to the feature vectors of multiple biometric maps to achieve weighted processing of the feature vectors of multiple biometric maps. For example, if the feature vectors of two biometric maps are p and q respectively, then the feature vector after feature weighting can be expressed as z=xp+yq, where z is the fused feature vector, x is the weight value of the feature vector p of one biometric map, and y is the weight value of the feature vector q of another biometric map.

[0119] In summary, the overall process of the continuous feature map feature extraction and fusion module for extracting and fusing multiple biological feature maps in the action sequence can be roughly referred to as Figure 6. The process may include: performing feature extraction processing on each of the multiple biometric graphs in the action sequence in a serial or parallel manner (such as performing feature extraction processing on each biometric graph using a deep learning model with feature extraction function), and extracting a feature vector from each biometric graph; the feature vector is used to represent part of the feature information of the biometric contained in the single biometric graph from the semantic dimension of the single biometric graph. Then, the feature vectors of each of the multiple biometric graphs in the action sequence are subjected to feature fusion (such as the fusion methods such as feature connection or feature weighting mentioned above) to generate a fused feature vector; the fused feature vector utilizes the complementarity (or authenticity, superposition) between the biometrics contained in the multiple consecutive biometric graphs in the action sequence, and can represent the comprehensive feature information of the biometrics, represent the feature information of all the biometric graphs in the action sequence, improve the performance and robustness of the feature recognition module, and enhance the ability to recognize and identify individual identities.

[0120] S304: Perform identity recognition on the biometric features based on the fused feature vector to obtain a recognition result.

[0121] After obtaining the fused feature vector of the biometric feature based on the above steps, the object to which the biometric feature belongs can be identified based on the fused feature vector to identify the object to which the biometric feature belongs, thereby realizing the identification of the individual's identity. The feature recognition module provided in the embodiment of the present application relies on the comparison and matching technology of biometric features, and uses the feature comparison algorithm to process and analyze the input fused feature vector to determine its similarity with the candidate biometric features already in the biometric feature data set (such as the database mentioned above), thereby realizing identity recognition and matching based on biometric features.

[0122] In summary, after the embodiment of the present application obtains a video stream by real-time and continuous acquisition of the target action performed by the user, the video stream can be further processed to obtain a more continuous action sequence. In this way, by extracting and fusing the features of multiple biometric graphs in the action sequence, more comprehensive and rich feature information of the biometrics can be obtained, and the perception of individual differences by the subsequent feature recognition model can be enhanced. In detail, ① by fusing the features of multiple consecutive biometric graphs, the impact of noise and changes caused by a single biometric can be reduced, and the robustness of the system can be improved (that is, when some biometric graphs are interfered with or of poor quality, the features of other biometric graphs can still make up for it and provide reliable identification information). ② The feature fusion of multiple consecutive biometric graphs can improve the one-to-one matching performance; specifically, the feature fusion of multiple biometric graphs can provide more comprehensive and accurate feature information, better capture the details and uniqueness of the biometrics, and improve the accuracy of subsequent biometric matching. ③ The feature fusion of multiple continuous biometric feature maps can also improve the one-to-many retrieval effect; specifically, in the one-to-many identity recognition process, the feature fusion of multiple continuous biometric feature maps can provide richer feature information, enhance the feature recognition module's ability to distinguish biometric features, improve the feature recognition module's retrieval effect in large-scale databases, and reduce the misrecognition rate. ④ Based on the continuous feature map acquisition technology of video streams, the embodiment of the present application supports dynamic biometric feature recognition; specifically, the feature fusion of multiple continuous biometric feature maps enables the feature recognition module to recognize dynamic biometric features, that is, in the process of continuous actions being performed on the biometric features, the feature recognition model can capture the temporal change pattern of the biometric features, thereby achieving accurate recognition of dynamic biometric features.

[0123] Figure 7 A schematic diagram showing a flow chart of another identity recognition method provided by an exemplary embodiment of the present application is shown; Figure 7 The identity recognition method shown may be executed by a computer device, and the identity recognition method may include but is not limited to steps S701-S705:

[0124] S701: Acquire an action sequence of biometric features.

[0125] It should be noted that the specific implementation process shown in step S701 can refer to the aforementioned Figure 3 The relevant description of the specific implementation process shown in step S301 in the illustrated embodiment is not repeated here.

[0126] As can be seen from the above description, the continuous multiple biometric graphs contained in the biometric action sequence are part of the initial biometric graphs screened from the video stream. In order to improve the image quality of the biometric graphs contained in the action sequence and increase the accuracy of identity recognition; the embodiment of the present application also supports image enhancement processing of the biometric graphs in the action sequence to improve the image quality, which is conducive to extracting more real and accurate feature information of the biometric based on high-quality biometric graphs, thereby improving the perception of personal identity.

[0127] Combine the following Figure 8 The process of optimizing the quality of biometric features in action sequences is introduced. Figure 8 As shown, first, after obtaining the action sequence of the biometric feature, an image quality detection is performed on each biometric feature image in the action sequence to obtain a quality detection result of each biometric feature image; if a resolution detection is performed on the biometric feature image, the lower the resolution of the biometric feature image, the lower the clarity of the biometric feature image; the embodiment of the present application does not limit the specific algorithm and process of image quality detection.

[0128] Then, based on the quality detection result of the biometric image, determine whether the biometric image needs to be enhanced; specifically, screen out the biometric images whose quality detection results do not meet the quality requirements from the action sequence, such as screening out some biometric images whose resolution is lower than the resolution threshold from the action sequence as the biometric images whose quality detection results do not meet the quality requirements.

[0129] Secondly, the biometric image whose quality inspection results do not meet the quality requirements is subjected to image enhancement processing to obtain the biometric image after image enhancement; and the biometric image whose quality inspection results meet the quality requirements is directly used for subsequent feature extraction. Among them, image enhancement can be simply understood as the process of making the originally unclear image clear or emphasizing the features of interest in the image, expanding the differences between the features of different objects in the image, improving the image quality and enriching the amount of information, so as to enhance the image interpretation and recognition effect. The embodiment of the present application does not limit the image enhancement algorithm used for image enhancement processing.

[0130] Exemplarily, the embodiment of the present application supports personalized selection of appropriate image enhancement algorithms for different biometric images according to the image processing requirements of each biometric image that requires image enhancement; through the matching of the image processing requirements of such biometric images and the image enhancement algorithm, the image quality of a single biometric image after image enhancement processing is improved. In a specific implementation, taking any biometric image in an action sequence whose quality detection result does not meet the quality requirements as an example, the process of image enhancement processing of the biometric image whose quality detection result does not meet the quality requirements may include but is not limited to: obtaining the image processing requirements of the target biometric image, and selecting an image enhancement algorithm that matches the image processing requirements; using the image processing algorithm that matches the image processing requirements to perform image enhancement processing on the target biometric image, and obtain the target biometric image after image enhancement processing. For example, if the image processing requirement of the target biometric image is to enhance the high-frequency information in the target biometric image, the image enhancement algorithm that matches the image processing requirement of the biometric image may include but is not limited to: an interpolation algorithm and a learning algorithm under a super-resolution algorithm. Among them, super-resolution technology is a technology that improves image resolution through image processing algorithms; it can restore clearer and more detailed high-resolution images from low-resolution images, thereby improving image quality and recognition accuracy. In detail, super-resolution technology usually involves two key principles: interpolation method and learning method. Interpolation method is a basic principle of super-resolution technology; by interpolating pixels of low-resolution images, the missing high-frequency details are inferred based on the information of neighboring pixels. Common interpolation methods include bilinear interpolation and bicubic interpolation. Learning method is a more advanced principle in super-resolution technology; a deep learning model is used to learn the mapping relationship between low-resolution images and high-resolution images. By training the model, it can learn a more accurate way to restore details, thereby generating high-quality high-resolution images. Of course, the image enhancement algorithm used for image enhancement of biometric features can include other algorithms (such as grayscale change algorithm or histogram, etc.) in addition to super-resolution algorithms, and there is no limitation on this.

[0131] Finally, the biometric action sequence is updated by using the enhanced biometric image to obtain an updated action sequence; the updated action sequence includes: the enhanced biometric image and the biometric image whose quality inspection result meets the quality requirements. Specifically, the biometric image whose quality inspection result does not meet the quality requirements is deleted from the biometric action sequence, and the corresponding enhanced biometric image is inserted into the action sequence according to the arrangement position of the biometric image whose quality inspection result does not meet the quality requirements in the action sequence to obtain an updated action sequence.

[0132] It can be seen that the embodiment of the present application uses an image enhancement algorithm (such as super-resolution technology) to process the collected biometric feature image to improve the details and clarity of the image, so that the enhanced biometric feature image is clearer and the lost details are restored, which is conducive to the subsequent reference to more accurate and detailed feature information about the biometric features from the high-quality biometric feature image, thereby improving the recognition performance of the biometric features.

[0133] S702: Perform feature extraction processing on each of the multiple biological feature maps to obtain a feature vector of each biological feature map.

[0134] S703: Fusing the feature vectors of each biological feature map to generate a fused feature vector.

[0135] It should be noted that the specific implementation process shown in steps S702-S703 can be found in the aforementioned Figure 3 The relevant description of the specific implementation process shown in steps S302-S303 in the illustrated embodiment is not repeated here.

[0136] S704: Obtain a biometric feature data set, and use a feature comparison algorithm to compare the comprehensive feature information indicated by the fused feature vector with each candidate feature information in the biometric feature data set, to obtain a feature matching result corresponding to each candidate feature information.

[0137] S705: Generate a recognition result based on the feature matching result corresponding to each candidate feature information.

[0138] In steps S704-S705, after obtaining the fused feature vector of multiple biometric feature images in the action sequence based on the previous steps, identity recognition can be performed on the biometric features from feature dimensions such as the morphology and texture of the biometric features based on the fused feature vector to achieve confirmation and authentication of the individual's identity. This relies on the comprehensive feature information of the biometric features indicated by the fused feature vector, making the identification of the individual's identity unique, stable and highly reliable.

[0139] In a specific implementation: (1) a biometric dataset may be obtained first, which includes candidate feature information of biometric features of different objects. In other words, the biometric dataset is a database that pre-stores the real biometric features of a large number of users; thus, when user identification is required, identification and matching can be performed based on the real database.

[0140] (2) Using a feature comparison algorithm, the comprehensive feature information indicated by the fused feature vector is compared with each candidate feature information in the biometric data set to obtain a feature matching result corresponding to each candidate feature information. The feature comparison algorithm may include but is not limited to: a trained deep learning model or other pattern recognition algorithm. The following takes the feature comparison algorithm including a similarity algorithm and a distance comparison algorithm as an example to exemplify the process of using a feature comparison algorithm to compare the comprehensive feature information indicated by the fused feature vector with each candidate feature information in the biometric data set to obtain a feature matching result corresponding to each candidate feature information.

[0141] In a specific implementation, on the one hand, the similarity of the comprehensive feature information indicated by the fused feature vector and each candidate feature information in the biometric data set can be calculated using a similarity algorithm from the similarity dimension of the image content, and the similarity comparison result corresponding to each candidate feature information can be obtained. On the other hand, from the shooting distance dimension during image acquisition, the distance comparison algorithm can be used to calculate the distance index of the first distance between the biometric corresponding to the fused feature vector and the acquisition device when it is acquired, and the second distance corresponding to each candidate feature information in the biometric data set, and the distance index result corresponding to each candidate feature information can be obtained; the distance index result is used to indicate the distance difference between the first distance and the second distance; the second distance corresponding to the candidate feature information refers to the distance between the biometric corresponding to the candidate feature information and the acquisition device when it is acquired.

[0142] Then, based on the similarity comparison result and the distance index result corresponding to each candidate feature information, a feature matching result corresponding to each candidate feature information is generated. Specifically, ① when the feature matching result is the weighted value of the similarity comparison result and the distance index result, the feature matching result satisfies the matching condition including: the weighted value is greater than the weighted threshold. For example, the feature matching result corresponding to candidate feature information 1 = the similarity comparison result corresponding to candidate feature information 1 * 20% + the distance index result corresponding to candidate feature information 1 * 80%. If the feature matching result is greater than the weighted threshold, it is determined that the feature matching result corresponding to the candidate feature information 1 satisfies the matching condition, that is, there is candidate feature information in the biometric data set whose feature matching result satisfies the matching condition. ② When the feature matching result includes the similarity comparison result and the distance index result, the feature matching result satisfies the matching condition including: the similarity comparison result is greater than the similarity threshold, and the distance index result is greater than the distance difference threshold. For example, the feature matching results corresponding to candidate feature information 1 include: the similarity comparison result corresponding to candidate feature information 1 and the distance index result corresponding to candidate feature information 1; if the similarity comparison result corresponding to candidate feature information 1 is greater than the similarity threshold, and the distance index result corresponding to candidate feature information 1 is greater than the distance difference threshold, then it is determined that the feature matching result corresponding to candidate feature information 1 meets the matching condition.

[0143] (3) Generate a recognition result based on the feature matching result corresponding to each candidate feature information. According to the feature matching result corresponding to each candidate feature information, it can be determined that the recognition result indicates that there is candidate feature information whose feature matching result satisfies the matching condition in the biometric data set, or that the recognition result indicates that there is no candidate feature information whose feature matching result satisfies the matching condition in the biometric data set. For example, assuming that the candidate feature information whose feature matching result satisfies the matching condition includes: the value of the feature matching result is greater than the matching threshold 95%, then if the biometric data set includes candidate feature information 1, candidate feature information 2 and candidate feature information 3, and the feature matching result between candidate feature information 1 and the comprehensive feature information indicated by the fused feature vector is 10%, the feature matching result between candidate feature information 2 and the comprehensive feature information indicated by the fused feature vector is 96%, and the feature matching result between candidate feature information 3 and the comprehensive feature information indicated by the fused feature vector is 0.1%, then it is determined that the object to which candidate feature information 2 whose feature matching result is greater than 95% belongs is the object for identity recognition for the current execution of the target action, and at this time, it is determined that there is candidate feature information whose feature matching result satisfies the matching condition in the biometric data set.

[0144] It is worth noting that, when the recognition result indicates that there is candidate feature information in the biometric data set whose feature matching result meets the matching conditions, the recognition result may also include object information of the object to which the biometric belongs. The object information included in the recognition result varies depending on the business scenario's requirements for the user's object information. For example: in a transportation ticket purchase scenario, the recognition result needs to include the account information, train information, and transaction information of the object to which the biometric belongs; for another example: in a security access control scenario, the recognition result needs to include the object name of the object to which the biometric belongs. The embodiments of the present application do not limit the type and quantity of object information specifically included in the recognition algorithm.

[0145] In the embodiment of the present application, it is possible to continuously collect multiple biometric feature graphs of the user in the process of performing the target action (such as palm swiping action / palm swiping behavior), and analyze the comprehensive and rich vector representation of the biometric features based on the multiple biometric feature graphs for identity recognition. Compared with identity recognition based on the feature vector of a single biometric feature graph, the accuracy and reliability of the biometric features used for identity recognition are effectively improved, thereby improving the accuracy of identity recognition. In addition, it supports image enhancement processing of some biometric feature graphs in the action sequence whose quality inspection results do not meet the quality requirements; this selective image enhancement processing of some biometric feature graphs can not only save image processing resources, but also ensure high-resolution images of the biometric feature graphs in the action sequence, thereby ensuring the reliability of subsequent feature extraction, feature fusion and recognition, thereby improving the recognition performance of personal identity.

[0146] The method of the embodiment of the present application is described in detail above. In order to facilitate better implementation of the above scheme of the embodiment of the present application, the device of the embodiment of the present application is provided below accordingly.

[0147] Fig. 9 A schematic diagram of the structure of an identity recognition device provided by an exemplary embodiment of the present application is shown; the identity recognition device can be used to perform Figure 3 or Figure 7 Some or all of the steps in the method embodiment shown. Fig. 9 , the identity recognition device comprises the following units:

[0148] An acquisition unit 901 is used to acquire an action sequence of a biometric feature; the action sequence includes a plurality of biometric feature graphs continuously collected for the biometric feature; and the feature changes between the biometric features included in any two adjacent biometric feature graphs in the plurality of biometric feature graphs are continuous;

[0149] A processing unit 902 is configured to perform feature extraction processing on each of multiple biometric images to obtain a feature vector for each biometric image; the feature vector of any biometric image represents sub-feature information of the biometric feature from the semantic dimension of the any biometric image;

[0150] The processing unit 902 is further configured to perform fusion processing on the feature vectors of each biometric image to generate a fused feature vector; the fused feature vector represents comprehensive feature information of the biometric feature from the semantic dimension of the action sequence;

[0151] The processing unit 902 is further configured to perform identity recognition on the biometric feature based on the fused feature vector to obtain a recognition result, and the recognition result indicates the object to which the biometric feature belongs.

[0152] In one implementation, when the acquisition unit 901 is configured to acquire the action sequence of the biometric feature, it is specifically configured to:

[0153] Continuously acquire the biometric feature of the target action being performed in the business scenario to obtain a video stream of the biometric feature; the video stream includes multiple initial biometric images acquired continuously, and each initial biometric image includes all or part of the features of the biometric feature;

[0154] Preprocess the multiple initial biometric images included in the video stream to obtain the action sequence of the biometric feature;

[0155] Wherein, the biometric image is a palm image, the biometric feature included in the biometric image is a palmprint feature, and the business scenario is a long-distance palm-sweeping scenario.

[0156] In one implementation, the video stream includes N initial biometric images, where N is an integer greater than 1; when the processing unit 902 is configured to preprocess the multiple initial biometric images included in the video stream to obtain the action sequence of the biometric feature, it is specifically configured to:

[0157] Determine whether the feature change between the biometric feature included in the (i + 1)-th initial biometric image in the video stream and the biometric feature included in the i-th initial biometric image in the video stream is continuous; 1 ≤ i < N, and i is an integer;

[0158] If it is continuous, add the (i + 1)-th initial biometric image to the action sequence of the biometric feature, and determine whether the (i + 1)-th initial biometric image is the last one among the N initial biometric images included in the video stream;

[0159] If the i+1th initial biometric feature map is not the last one of the N initial biometric feature maps contained in the video stream, then continue to determine whether the feature changes between the biometric features contained in the i+2th initial biometric feature map and the biometric features contained in the i+1th initial biometric feature map are continuous, until a candidate biometric feature map is detected in the video stream;

[0160] The biometric feature map includes: the last one of the N initial biometric feature maps contained in the video stream, or an initial biometric feature whose feature change between the contained biometric feature map and the biometric feature contained in the adjacent previous biometric feature map is not continuous.

[0161] In one implementation, the processing unit 902 is further configured to:

[0162] If there is no continuity, then determine the action sequence of the biometric feature; the action sequence includes i initial biometric feature images in the video stream, and the i initial biometric feature images are: the initial biometric feature image before the i+1th initial biometric feature image in the video stream.

[0163] In one implementation, if a deep learning model is used to determine whether the feature changes between the biometric features included in the i+1th initial biometric feature graph and the biometric features included in the i-th initial biometric feature graph are continuous, the processing unit 902 is further used to:

[0164] Add the i+1th initial biometric feature map to the interference data set;

[0165] Among them, the interference data set includes: an initial biometric feature map in which the feature changes between the biometric features included in the video stream and the previous adjacent initial biometric feature map are not continuous; the interference data set serves as interference data for the deep learning model during the model training process.

[0166] In one implementation, the processing unit 902 is further configured to:

[0167] Performing image quality detection on each biometric feature image in the action sequence to obtain a quality detection result of each biometric feature image;

[0168] Filter out the biometric graphs whose quality inspection results do not meet the quality requirements from the action sequence;

[0169] Performing image enhancement processing on the biometric image whose quality inspection result does not meet the quality requirements to obtain an image-enhanced biometric image;

[0170] The biometric action sequence is updated by using the biometric image after image enhancement to obtain an updated action sequence; the updated action sequence includes: the biometric image after image enhancement and the biometric image whose quality detection result meets the quality requirements.

[0171] In one implementation, the biometric image whose quality inspection result does not meet the quality requirements is represented as a target biometric image; the processing unit 902 is used to perform image enhancement processing on the biometric image whose quality inspection result does not meet the quality requirements, and when the biometric image after image enhancement is obtained, it is specifically used to:

[0172] Obtain the image processing requirements of the target biometric image and select an image enhancement algorithm that matches the image processing requirements;

[0173] Using an image processing algorithm that matches the image processing requirements, the target biological characteristic image is subjected to image enhancement processing to obtain the target biological characteristic image after image enhancement processing;

[0174] Among them, if the image processing requirement is to enhance the high-frequency information in the target biological feature map, the image enhancement algorithms that match the image processing requirements include: interpolation algorithms and learning algorithms under super-resolution algorithms.

[0175] In one implementation, the processing unit 902 is used to perform identity recognition on the biometric feature based on the fused feature vector, and when obtaining the recognition result, is specifically used to:

[0176] Acquire a biometric feature data set, wherein the biometric feature data set includes candidate feature information of biometric features of different objects;

[0177] Using a feature comparison algorithm, the comprehensive feature information indicated by the fused feature vector is compared with each candidate feature information in the biometric feature data set to obtain a feature matching result corresponding to each candidate feature information;

[0178] Generate a recognition result based on the feature matching result corresponding to each candidate feature information;

[0179] The recognition result indicates that there is candidate feature information in the biometric data set whose feature matching result satisfies the matching condition, or the recognition result indicates that there is no candidate feature information in the biometric data set whose feature matching result satisfies the matching condition.

[0180] In one implementation, the feature comparison algorithm includes a similarity algorithm and a distance comparison algorithm; the processing unit 902 is used to use the feature comparison algorithm to compare the comprehensive feature information indicated by the fused feature vector with each candidate feature information in the biometric feature data set, and obtain the feature matching result corresponding to each candidate feature information, specifically for:

[0181] Using a similarity algorithm, respectively, similarity calculations are performed on the comprehensive feature information indicated by the fused feature vector and each candidate feature information in the biometric feature data set to obtain a similarity comparison result corresponding to each candidate feature information; and,

[0182] By using a distance comparison algorithm, a distance index is calculated for each of the first distances between the biometric feature corresponding to the fused feature vector and the collection device when it is collected, and the second distance corresponding to each candidate feature information in the biometric feature data set, to obtain a distance index result corresponding to each candidate feature information; the distance index result is used to indicate the distance difference between the first distance and the second distance; the second distance corresponding to the candidate feature information refers to the distance between the biometric feature corresponding to the candidate feature information and the collection device when it is collected;

[0183] Based on the similarity comparison result and the distance index result corresponding to each candidate feature information, a feature matching result corresponding to each candidate feature information is generated.

[0184] In one implementation, when the feature matching result is a weighted value of a similarity comparison result and a distance index result, the feature matching result satisfies the matching condition including: the weighted value is greater than a weighted threshold;

[0185] When the feature matching result includes a similarity comparison result and a distance index result, the feature matching result satisfies the matching condition including: the similarity comparison result is greater than a similarity threshold, and the distance index result is greater than a distance difference threshold.

[0186] According to one embodiment of the present application, Fig. 9 The various units in the identity recognition device shown can be individually or completely combined into one or several other units to form, or one (some) of the units can be further divided into multiple functionally smaller units to form, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In practical applications, the function of one unit can also be implemented by multiple units, or the functions of multiple units can be implemented by one unit. In other embodiments of the present application, the identity recognition device may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units. According to another embodiment of the present application, it can be executed by running on a general computing device such as a computer that includes processing elements and storage elements such as a central processing unit (CPU), a random access storage medium (RAM), and a read-only storage medium (ROM). Figure 2 and Figure 7 A computer program (including program code) for each step involved in the corresponding method shown in FIG. Fig. 9The identity recognition device shown in the embodiment of the present application is implemented by the identity recognition method. The computer program can be recorded on a computer-readable recording medium, for example, and loaded into the above-mentioned computing device through the computer-readable recording medium and run therein.

[0187] In the embodiment of the present application, it is supported to continuously collect the biometric features of the user to obtain the action sequence of the biometric features; wherein the action sequence includes multiple biometric feature graphs obtained by continuous collection of the biometric features, and the feature changes between the biometric features contained in any two adjacent biometric feature graphs in the multiple biometric feature graphs are continuous. In this way, feature extraction processing can be performed on multiple biometric feature graphs in the action sequence to obtain a feature vector of each biometric feature graph; the feature vector of any biometric feature graph expresses the sub-feature information of the biometric feature from the semantic dimension of the biometric feature. For example, if the biometric feature is a palm, and the palm in the biometric feature graph is tilted, then the feature vector in the biometric feature graph represents the palm print characteristics when the palm is tilted. Furthermore, the feature vectors of each biometric feature graph are fused to generate a fused feature vector, which represents the comprehensive feature information of the biometric feature from the semantic dimension of the action sequence. Finally, identity recognition is performed based on the fused feature information of the biometric feature to obtain an identification result indicating the object to which the biometric feature belongs. It can be seen from the above scheme that the embodiment of the present application provides a new biometric collection and identification scheme, which can continuously collect multiple biometric graphs of the user when performing the target action (such as palm swiping action / palm swiping behavior), and analyze the comprehensive and rich vector representation of the biometrics based on multiple biometric graphs for identity identification; compared with identity identification based on the feature vector of a single biometric graph, it effectively improves the accuracy and reliability of the biometrics used for identity identification, thereby improving the accuracy of identity identification.

[0188] Fig.10 FIG. 1 shows a schematic diagram of a computer device provided by an exemplary embodiment of the present application. Fig.10, the computer device includes a processor 1001, a communication interface 1002 and a computer-readable storage medium 1003. The processor 1001, the communication interface 1002 and the computer-readable storage medium 1003 can be connected via a bus or other means. The communication interface 1002 is used to receive and send data. The computer-readable storage medium 1003 can be stored in the memory of the computer device, the computer-readable storage medium 1003 is used to store a computer program, the computer program includes program instructions, and the processor 1001 is used to execute the program instructions stored in the computer-readable storage medium 1003. The processor 1001 (or CPU (Central Processing Unit)) is the computing core and control core of the computer device, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function.

[0189] The embodiment of the present application also provides a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space that stores the processing system of the computer device. In addition, one or more instructions suitable for being loaded and executed by the processor 1001 are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory, or a non-volatile memory, such as at least one disk storage; optionally, it can also be at least one computer-readable storage medium located away from the aforementioned processor.

[0190] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor 1001 loads and executes the one or more instructions stored in the computer-readable storage medium to implement the corresponding steps in the above-mentioned identity recognition method embodiment; in a specific implementation, the one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and the following steps are executed:

[0191] An action sequence for acquiring biometric features; the action sequence includes a plurality of biometric feature graphs continuously collected for the biometric features; and the feature changes between the biometric features included in any two adjacent biometric feature graphs among the plurality of biometric feature graphs are continuous;

[0192] Feature extraction processing is performed on each of multiple biometric images to obtain a feature vector for each biometric image; the feature vector of any biometric image represents sub-feature information of the biometric from the semantic dimension of the biometric image;

[0193] The feature vectors of each biometric image are fused to generate a fused feature vector; the fused feature vector represents comprehensive feature information of the biometric from the semantic dimension of the action sequence;

[0194] Based on the fused feature vector, identity recognition of the biometric is performed to obtain a recognition result, and the recognition result indicates the object to which the biometric belongs.

[0195] In one implementation, when one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and perform the action sequence of acquiring biometrics, the following steps are specifically executed:

[0196] Continuously collect the biometrics of the target action to be performed in the business scenario to obtain a video stream of the biometrics; the video stream includes multiple initial biometric images collected continuously, and each initial biometric image includes all or part of the features of the biometric;

[0197] Preprocess the multiple initial biometric images included in the video stream to obtain an action sequence of the biometrics;

[0198] Wherein, the biometric image is a palm image, the biometric included in the biometric image is a palmprint feature, and the business scenario is a long-distance palm-sweeping scenario.

[0199] In one implementation, the video stream includes N initial biometric images, where N is an integer greater than 1; when one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and perform preprocessing on the multiple initial biometric images included in the video stream to obtain an action sequence of the biometrics, the following steps are specifically executed:

[0200] Determine whether the feature change between the biometric included in the (i + 1)-th initial biometric image and the biometric included in the i-th initial biometric image in the video stream is continuous; 1 ≤ i < N, and i is an integer;

[0201] If it is continuous, add the (i + 1)-th initial biometric image to the action sequence of the biometrics, and determine whether the (i + 1)-th initial biometric image is the last one among the N initial biometric images included in the video stream;

[0202] If the i+1th initial biometric feature map is not the last one of the N initial biometric feature maps contained in the video stream, then continue to determine whether the feature changes between the biometric features contained in the i+2th initial biometric feature map and the biometric features contained in the i+1th initial biometric feature map are continuous, until a candidate biometric feature map is detected in the video stream;

[0203] The biometric feature map includes: the last one of the N initial biometric feature maps contained in the video stream, or an initial biometric feature whose feature change between the contained biometric feature map and the biometric feature contained in the adjacent previous biometric feature map is not continuous.

[0204] In one implementation, one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and further perform the following steps:

[0205] If there is no continuity, then determine the action sequence of the biometric feature; the action sequence includes i initial biometric feature images in the video stream, and the i initial biometric feature images are: the initial biometric feature image before the i+1th initial biometric feature image in the video stream.

[0206] In one implementation, if a deep learning model is used to implement the judgment of whether the feature changes between the biometric features included in the i+1th initial biometric feature map and the biometric features included in the i-th initial biometric feature map are continuous, one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and the following steps are also executed:

[0207] Add the i+1th initial biometric feature map to the interference data set;

[0208] Among them, the interference data set includes: an initial biometric feature map in which the feature changes between the biometric features included in the video stream and the previous adjacent initial biometric feature map are not continuous; the interference data set serves as interference data for the deep learning model during the model training process.

[0209] In one implementation, one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and further perform the following steps:

[0210] Performing image quality detection on each biometric feature image in the action sequence to obtain a quality detection result of each biometric feature image;

[0211] Filter out the biometric graphs whose quality inspection results do not meet the quality requirements from the action sequence;

[0212] Performing image enhancement processing on the biometric image whose quality inspection result does not meet the quality requirements to obtain an image-enhanced biometric image;

[0213] The biometric action sequence is updated by using the biometric image after image enhancement to obtain an updated action sequence; the updated action sequence includes: the biometric image after image enhancement and the biometric image whose quality detection result meets the quality requirements.

[0214] In one implementation, the biometric image whose quality inspection result does not meet the quality requirements is represented as a target biometric image; one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and when the processor 1001 performs image enhancement processing on the biometric image whose quality inspection result does not meet the quality requirements to obtain the biometric image after image enhancement, the following steps are specifically performed:

[0215] Obtain the image processing requirements of the target biometric image and select an image enhancement algorithm that matches the image processing requirements;

[0216] Using an image processing algorithm that matches the image processing requirements, the target biological characteristic image is subjected to image enhancement processing to obtain the target biological characteristic image after image enhancement processing;

[0217] Among them, if the image processing requirement is to enhance the high-frequency information in the target biological feature map, the image enhancement algorithms that match the image processing requirements include: interpolation algorithms and learning algorithms under super-resolution algorithms.

[0218] In one implementation, one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and when performing identity recognition based on the fused feature vector for the biometric feature to obtain the recognition result, the following steps are specifically performed:

[0219] Acquire a biometric feature data set, wherein the biometric feature data set includes candidate feature information of biometric features of different objects;

[0220] Using a feature comparison algorithm, the comprehensive feature information indicated by the fused feature vector is compared with each candidate feature information in the biometric feature data set to obtain a feature matching result corresponding to each candidate feature information;

[0221] Generate a recognition result based on the feature matching result corresponding to each candidate feature information;

[0222] The recognition result indicates that there is candidate feature information in the biometric data set whose feature matching result satisfies the matching condition, or the recognition result indicates that there is no candidate feature information in the biometric data set whose feature matching result satisfies the matching condition.

[0223] In one implementation, the feature comparison algorithm includes a similarity algorithm and a distance comparison algorithm; one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and when executing the feature comparison algorithm, the comprehensive feature information indicated by the fused feature vector is compared with each candidate feature information in the biometric feature data set, and the feature matching result corresponding to each candidate feature information is obtained, and the following steps are specifically performed:

[0224] Using a similarity algorithm, respectively, similarity calculations are performed on the comprehensive feature information indicated by the fused feature vector and each candidate feature information in the biometric feature data set to obtain a similarity comparison result corresponding to each candidate feature information; and,

[0225] By using a distance comparison algorithm, a distance index is calculated for each of the first distances between the biometric feature corresponding to the fused feature vector and the collection device when it is collected, and the second distance corresponding to each candidate feature information in the biometric feature data set, to obtain a distance index result corresponding to each candidate feature information; the distance index result is used to indicate the distance difference between the first distance and the second distance; the second distance corresponding to the candidate feature information refers to the distance between the biometric feature corresponding to the candidate feature information and the collection device when it is collected;

[0226] Based on the similarity comparison result and the distance index result corresponding to each candidate feature information, a feature matching result corresponding to each candidate feature information is generated.

[0227] In one implementation, when the feature matching result is a weighted value of a similarity comparison result and a distance index result, the feature matching result satisfies the matching condition including: the weighted value is greater than a weighted threshold;

[0228] When the feature matching result includes a similarity comparison result and a distance index result, the feature matching result satisfies the matching condition including: the similarity comparison result is greater than a similarity threshold, and the distance index result is greater than a distance difference threshold.

[0229] Based on the same inventive concept, the principles and beneficial effects of solving the problems provided by the computer device in the embodiment of the present application are similar to the principles and beneficial effects of solving the problems provided by the identity recognition method in the method embodiment of the present application. Please refer to the principles and beneficial effects of the implementation of the method. For the sake of concise description, they will not be repeated here.

[0230] The embodiment of the present application also provides a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device performs the above-mentioned identity recognition method.

[0231] A person skilled in the art can appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0232] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loading and executing computer program instructions on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instruction can be stored in a computer-readable storage medium or transmitted by a computer-readable storage medium. The computer instruction can be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (for example, coaxial cable, optical fiber, digital line (DSL)) or wireless (for example, infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server, a data center, etc. that contains one or more available media integration. Available media can be magnetic media (for example, floppy disk, hard disk, tape), optical media (for example, DVD), or semiconductor media (for example, solid-state drive (Solid State Disk, SSD)) and the like.

[0233] The above description is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technical object familiar with the technical field can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An identity recognition method, characterized in that: Including: An action sequence for acquiring a biometric; the action sequence includes a plurality of biometric maps continuously acquired for the biometric; the feature changes between any two adjacent biometric maps among the plurality of biometric maps are continuous; Performing feature extraction processing on each biometric map in the plurality of biometric maps to obtain a feature vector of each biometric map; the feature vector of any biometric map represents sub-feature information of the biometric from the semantic dimension of the any biometric map; Performing fusion processing on the feature vectors of each biometric map to generate a fused feature vector; the The fused feature vector represents comprehensive feature information of the biometric from the semantic dimension of the action sequence; Performing identity recognition on the biometric based on the fused feature vector to obtain a recognition result, and the recognition result indicates the object to which the biometric belongs.

2. The method according to claim 1, characterized in that The action sequence for acquiring a biometric includes: Continuously acquiring the biometric of the target action being performed in a business scenario to obtain a video stream of the biometric; the video stream includes a plurality of initial biometric maps continuously acquired, and each initial biometric map includes all or part of the features of the biometric; Performing preprocessing on the plurality of initial biometric maps included in the video stream to obtain the action sequence of the biometric; Wherein, the biometric map is a palm map, the biometric included in the biometric map is a palmprint feature, and the business scenario is a long-distance palm-sweeping scenario.

3. The method according to claim 2, characterized in that The video stream includes N initial biometric maps, and N is an integer greater than 1; the performing preprocessing on the plurality of initial biometric maps included in the video stream to obtain the action sequence of the biometric includes: Judging whether the feature change between the biometric included in the (i + 1)-th initial biometric map in the video stream and the biometric included in the i-th initial biometric map in the video stream is continuous; 1 ≤ i < N, and i is an integer; If it is continuous, adding the (i + 1)-th initial biometric map to the action sequence of the biometric, and judging whether the (i + 1)-th initial biometric map is the last one among the N initial biometric maps included in the video stream; If the (i + 1)-th initial biometric map is not the last one among the N initial biometric maps included in the video stream, continue to judge whether the feature change between the biometric included in the (i + 2)-th initial biometric map and the biometric included in the (i + 1)-th initial biometric map is continuous until a candidate biometric map in the video stream is detected; Wherein, the candidate biometric map includes: the last one among the N initial biometric maps included in the video stream, or, an initial biometric whose feature change between the included biometric map and the biometric included in the adjacent previous biometric map is not continuous.

4. The method according to claim 3, characterized in that The method further includes: If there is no continuity, determine the action sequence of the biometric; the action sequence includes i initial biometric characteristic images in the video stream, and the i initial biometric characteristic images are: the initial biometric characteristic image in the video stream that is located before the i+1th initial biometric characteristic image.

5. The method according to claim 4, characterized in that If a deep learning model is used to implement the judgment of whether the feature changes between the biometric features included in the i+1th initial biometric feature graph and the biometric features included in the i-th initial biometric feature graph are continuous, the method further includes: Adding the i+1th initial biometric feature map to the interference data set; Among them, the interference data set includes: an initial biometric feature map in which the feature changes between the biometric features included in the video stream and the previous adjacent initial biometric feature map are not continuous; the interference data set serves as interference data for the deep learning model during the model training process.

6. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: Performing image quality detection on each biometric feature image in the action sequence to obtain a quality detection result of each biometric feature image; Screening out biometric feature graphs whose quality inspection results do not meet quality requirements from the action sequence; Performing image enhancement processing on the biometric image whose quality inspection result does not meet the quality requirements to obtain an image-enhanced biometric image; The biometric image after image enhancement is used to update the action sequence of the biometric feature to obtain an updated action sequence; the updated action sequence includes: the biometric image after image enhancement and the biometric image whose quality detection result meets the quality requirements.

7. The method according to claim 6, characterized in that The biometric image whose quality inspection result does not meet the quality requirements is represented as a target biometric image; the image enhancement processing is performed on the biometric image whose quality inspection result does not meet the quality requirements to obtain the biometric image after image enhancement, including: Obtaining image processing requirements of the target biometric image, and selecting an image enhancement algorithm that matches the image processing requirements; Using an image processing algorithm that matches the image processing requirement, the target biological characteristic image is subjected to image enhancement processing to obtain the target biological characteristic image after image enhancement processing; Among them, if the image processing requirement is to enhance the high-frequency information in the target biological feature image, the image enhancement algorithms matching the image processing requirement include: an interpolation algorithm and a learning algorithm under a super-resolution algorithm.

8. The method according to claim 1, characterized in that The performing identity recognition on the biometric feature based on the fused feature vector to obtain a recognition result includes: Acquire a biometric feature data set, wherein the biometric feature data set includes candidate feature information of biometric features of different objects; Using a feature comparison algorithm, respectively compare the comprehensive feature information indicated by the fused feature vector with each candidate feature information in the biometric feature data set to obtain a feature matching result corresponding to each candidate feature information; Generate a recognition result based on the feature matching result corresponding to each candidate feature information; The recognition result indicates that there is candidate feature information in the biometric feature data set whose feature matching result satisfies the matching condition, or the recognition result indicates that there is no candidate feature information in the biometric feature data set whose feature matching result satisfies the matching condition.

9. The method according to claim 8, characterized in that The feature comparison algorithm includes the similarity algorithm and the distance comparison algorithm; the feature comparison algorithm is used to compare the comprehensive feature information indicated by the fused feature vector with each candidate feature information in the biometric feature data set, respectively, to obtain a feature matching result corresponding to each candidate feature information, including: Using the similarity algorithm, respectively, calculating the similarity of the comprehensive feature information indicated by the fused feature vector and each candidate feature information in the biometric feature data set to obtain a similarity comparison result corresponding to each candidate feature information; and, Using the distance comparison algorithm, respectively, a distance index calculation is performed on a first distance between the biometric feature corresponding to the fused feature vector and a collection device when it is collected, and a second distance corresponding to each candidate feature information in the biometric feature data set, to obtain a distance index result corresponding to each candidate feature information; the distance index result is used to indicate the distance difference between the first distance and the second distance; the second distance corresponding to the candidate feature information refers to the distance between the biometric feature corresponding to the candidate feature information and the collection device when it is collected; Based on the similarity comparison result and the distance index result corresponding to each candidate feature information, a feature matching result corresponding to each candidate feature information is generated.

10. The method according to claim 9, characterized in that When the feature matching result is a weighted value of the similarity comparison result and the distance index result, the feature matching result satisfies the matching condition including: the weighted value is greater than a weighted threshold; When the feature matching result includes the similarity comparison result and the distance index result, the feature matching result satisfies the matching condition including: the similarity comparison result is greater than a similarity threshold, and the distance index result is greater than a distance difference threshold.

11. An identity recognition device, characterized in that: include: An acquisition unit, configured to acquire an action sequence of a biometric feature; the action sequence includes a plurality of biometric feature graphs continuously collected for the biometric feature; and the feature changes between the biometric features included in any two adjacent biometric feature graphs among the plurality of biometric feature graphs are continuous; A processing unit, configured to perform feature extraction processing on each of the plurality of biometric feature graphs to obtain a feature vector of each biometric feature graph; the feature vector of any biometric feature graph represents sub-feature information of the biometric feature from a semantic dimension of the biometric feature graph; The processing unit is further used to fuse the feature vectors of each biometric feature image to generate a fused feature vector; the fused feature vector represents the comprehensive feature information of the biometric feature from the semantic dimension of the action sequence; The processing unit is further used to perform identity recognition on the biometric feature based on the fused feature vector to obtain a recognition result, wherein the recognition result indicates the object to which the biometric feature belongs.

12. A computer device, characterized in that: a processor adapted to execute a computer program; A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by the processor, the identity recognition method according to any one of claims 1 to 10 is implemented.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the identity recognition method according to any one of claims 1 to 10.

14. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the identity recognition method according to any one of claims 1 to 10 is implemented.

Images