Method, device, equipment and storage medium for feature extraction of biological texture information
Through the feature extraction method of biological texture information, combined with the fusion of global and local features, and using feature extraction and fusion networks, the problem of difficulty in identifying high-similar biological texture information such as twins is solved, and the recognition accuracy is improved.
Patent Information
- Application Number
- CN202210122885.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-02-09
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-02-09
AI Technical Summary
When existing biotexture information recognition technologies face biological lines with high similarity, their recognition ability, especially twins, has significantly decreased.
A feature extraction method for biological texture information is adopted. By obtaining biological texture information, the global texture features and local texture features are extracted and fusion, and the feature extraction network and feature fusion network are used, combined with hidden layer feature extraction and self-attention mechanism processing. and hollow convolution blocks to generate fusion texture features to improve recognition accuracy.
It improves attention to local details in biological lines, improves the ability of pattern features to describe palm lines information, and ensures the recognition accuracy when facing high similarity biological lines information.
Smart Images

Figure CN116631012B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of identity recognition technology, and in particular to a method, device, equipment and storage medium for extracting features of biological pattern information. Background Art
[0002] With the development of computer technology, the application scenarios of biometric pattern information recognition technology continue to expand, and it is widely used in attendance records, mobile payments and other scenarios.
[0003] In related technologies, deep features of bio-pattern information are usually extracted, and high-dimensional bio-pattern information is reduced in dimension during feature extraction to obtain low-dimensional bio-pattern features to describe the bio-pattern information, and the bio-pattern information is identified by comparing the bio-pattern features.
[0004] However, for biometric pattern information with a high degree of similarity, such as that between twins, the recognition ability decreases significantly. Summary of the invention
[0005] The present application provides a method, device, equipment and storage medium for extracting features of biological texture information. The technical solution is as follows:
[0006] According to one aspect of the present application, a method for extracting features of biological texture information is provided, the method comprising:
[0007] Acquire biological pattern information, where the biological pattern information is used to describe the biological pattern of the first object;
[0008] Performing feature extraction processing on the biological texture information to obtain global texture features and local texture features, wherein the global texture features are used to describe global features of the biological texture information of the first object, and the local texture features are used to describe local features of the biological texture information of the first object;
[0009] The global texture features and the local texture features of the first object are fused to obtain fused texture features of the first object.
[0010] According to another aspect of the present application, a method for extracting features of biological pattern information is provided, the method being performed by a biological pattern feature extraction model, the biological pattern feature extraction model comprising a feature extraction network and a feature fusion network;
[0011] The method comprises:
[0012] Acquire biological pattern information, where the biological pattern information is used to describe the biological pattern of the first object;
[0013] Calling the feature extraction network to perform feature extraction processing on the biological texture information to obtain global texture features and local texture features, wherein the global texture features are used to describe the global features of the biological texture information of the first object, and the local texture features are used to describe the local features of the biological texture information of the first object;
[0014] The feature fusion network is called to fuse the global texture features and the local texture features of the first object to obtain fused texture features of the first object.
[0015] According to another aspect of the present application, a feature extraction device for biological pattern information is provided, the device comprising:
[0016] An acquisition module, used for acquiring biological pattern information, where the biological pattern information is used for describing the biological pattern of the first object;
[0017] an extraction module, configured to perform feature extraction processing on the biological texture information to obtain global texture features and local texture features, wherein the global texture features are used to describe the global features of the biological texture information of the first object, and the local texture features are used to describe the local features of the biological texture information of the first object;
[0018] A fusion module is used to fuse the global texture features and the local texture features of the first object to obtain fused texture features of the first object.
[0019] In an optional design of the application, the extraction module is also used for:
[0020] Performing hidden layer feature extraction on the biometric pattern information to obtain texture hidden layer features of the first object, wherein the texture hidden layer features are used to describe the hidden layer representation of the target palmprint information;
[0021] Extracting the texture hidden layer features by a first feature extraction method to obtain the global texture features of the first object, wherein the first feature extraction method is a feature extraction method based on reducing feature dimensions;
[0022] The texture hidden layer features are extracted by a second feature extraction method to obtain the local texture features of the first object, wherein the second feature extraction method is a feature extraction method based on increasing the feature receptive field.
[0023] In an optional design of the application, the extraction module is also used for:
[0024] Processing the texture hidden layer features, respectively using n second feature extraction methods to obtain n local sub-features of the first object, where n is a positive integer, and the feature receptive fields of the n second feature extraction methods are different;
[0025] The n local sub-features are spliced to obtain the local texture feature of the first object.
[0026] In an optional design of the application, the device further comprises:
[0027] A processing module, used for performing a self-attention mechanism on the local texture features to obtain updated local texture features;
[0028] The fusion module is further used to fuse the global texture features and the updated local texture features of the first object to obtain fused texture features of the first object.
[0029] In an optional design of the application, the self-attention mechanism processing includes activation processing and regularization processing;
[0030] The processing module is further used for:
[0031] Performing the activation process on the local texture feature to obtain a second activation feature;
[0032] Performing the regularization process on the local texture feature to obtain a second regularized feature;
[0033] The second activation feature and the second regular feature are multiplied to obtain the updated local texture feature of the first object.
[0034] In an optional design of the application, the device further comprises:
[0035] The processing module is used to pre-process the biometric information to obtain updated biometric information; the pre-processing method includes but is not limited to at least one of the following:
[0036] Performing interpolation processing on the biological texture information;
[0037] intercepting a region of interest of the biometric pattern information;
[0038] The extraction module is also used to perform feature extraction processing on the updated biological texture information to obtain global texture features and local texture features.
[0039] In an optional design of the application, the device further comprises:
[0040] The judgment module is used to compare the fused texture features of the first object with the sample texture features to obtain the identity judgment result of the first object, and the sample texture features are used to describe the biological texture information features of the target object.
[0041] In an optional design of the application, the judgment module is also used for:
[0042] Calculating the cosine similarity between the fused texture feature and the sample texture feature;
[0043] When the cosine similarity satisfies an identity judgment condition, obtaining the identity judgment result of the first object is the target object;
[0044] When the cosine similarity does not satisfy the identity judgment condition, the identity judgment result obtained for the first object is not the target object.
[0045] According to another aspect of the present application, there is provided a feature extraction device for biometric pattern information, the device being executed by a biometric pattern feature extraction model, the biometric pattern feature extraction model comprising a feature extraction network and a feature fusion network;
[0046] The device comprises:
[0047] An acquisition module, used for acquiring biological pattern information, where the biological pattern information is used for describing the biological pattern of the first object;
[0048] an extraction module, used to call the feature extraction network to perform feature extraction processing on the biological texture information to obtain global texture features and local texture features, wherein the global texture features are used to describe the global features of the biological texture information of the first object, and the local texture features are used to describe the local features of the biological texture information of the first object;
[0049] A fusion module is used to call the feature fusion network to fuse the global texture features and the local texture features of the first object to obtain a fused texture feature of the first object.
[0050] In an optional design of the application, the feature extraction network includes: a first residual convolution block, a second residual convolution block and a dilated convolution block;
[0051] The extraction module is also used for:
[0052] Calling the first residual convolution block to perform hidden layer feature extraction on the biometric pattern information to obtain texture hidden layer features of the first object, wherein the texture hidden layer features are used to describe the hidden layer representation of the target palmprint information;
[0053] Calling the second residual convolution block to extract the texture hidden layer features by a first feature extraction method to obtain the global texture features of the first object, wherein the first feature extraction method is a feature extraction method based on reducing feature dimensions;
[0054] The atrous convolution block is called to extract the texture hidden layer features through a second feature extraction method to obtain the local texture features of the first object, wherein the second feature extraction method is a feature extraction method based on increasing the feature receptive field.
[0055] In an optional design of the application, the atrous convolution block includes n atrous convolution layers;
[0056] The extraction module is also used for:
[0057] Calling the n atrous convolution layers to process the texture hidden layer features, respectively obtaining n local sub-features of the first object through n second feature extraction methods, where n is a positive integer, and the n second feature extraction methods have different feature receptive fields, which are determined based on the convolution kernel scale of the atrous convolution layer;
[0058] The n local sub-features are spliced to obtain the local texture feature of the first object.
[0059] In an optional design of the application, the biometric pattern feature extraction model is trained in the following manner:
[0060] Acquire sample texture information and sample texture features, wherein the sample texture information is used to describe the biological texture of the sample object, and the sample texture features are biological texture features corresponding to the sample texture information of the sample object;
[0061] Calling the initial biological texture feature extraction model to process the sample texture information and output the predicted fused texture features of the sample object;
[0062] Comparing the predicted fusion texture features with the sample texture features to obtain a feature prediction error;
[0063] The feature prediction error is used to perform backward error propagation training on the initial biometric pattern feature extraction model to obtain the biometric pattern feature extraction model.
[0064] In an optional design of the application, the biometric pattern feature extraction model is trained in the following manner:
[0065] Acquire sample texture information and a classification label of the sample texture information, wherein the sample texture information is used to describe the biological texture of the sample object, and the classification label is used to indicate the sample object;
[0066] Calling the initial biological texture feature extraction model to process the sample texture information and output the predicted fused texture features of the sample object;
[0067] Performing label classification processing on the predicted fused texture features to obtain a predicted label of the sample object;
[0068] Comparing the predicted label with the classification label to obtain a label prediction error;
[0069] The label prediction error is used to perform backward error propagation training on the initial biometric pattern feature extraction model to obtain the biometric pattern feature extraction model.
[0070] According to another aspect of the present application, a computer device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the feature extraction method of biometric pattern information as described above.
[0071] According to another aspect of the present application, a computer-readable storage medium is provided, wherein at least one instruction, at least one program, a code set or an instruction set is stored in the computer-readable storage medium, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the feature extraction method of biological pattern information as described above.
[0072] According to another aspect of the present application, a computer program product or a computer program is provided, wherein the computer program product or the computer program comprises computer instructions, wherein the computer instructions are stored in a computer-readable storage medium, and a processor reads and executes the computer instructions from the computer-readable storage medium to implement the feature extraction method of the biological pattern information as described above.
[0073] The beneficial effects of the technical solution provided by this application include at least:
[0074] By extracting features from bio-print information to obtain local print features, the focus on local detail features in bio-prints is increased; by fusing local print features with global print features, the ability of print features to describe palmprint information is enhanced; and by expanding the dimension of print features to describe palmprint information locally, the recognition accuracy is ensured when facing bio-print information with high similarity. BRIEF DESCRIPTION OF THE DRAWINGS
[0075] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments 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 creative work.
[0076] Figure 1 is a schematic diagram of a computer system provided by an exemplary embodiment of the present application;
[0077] Figure 2 is a flow chart of a method for extracting features of biological texture information provided by an exemplary embodiment of the present application;
[0078] Figure 3 is a flow chart of a method for extracting features of biological texture information provided by an exemplary embodiment of the present application;
[0079] Figure 4 is a schematic diagram of a dilated convolution kernel provided by an exemplary embodiment of the present application;
[0080] Figure 5 is a flow chart of a method for extracting features of biological texture information provided by an exemplary embodiment of the present application;
[0081] Figure 6 is a schematic diagram of a ReLU function and a Softplus function provided by an exemplary embodiment of the present application;
[0082] Figure 7 is a flow chart of a method for extracting features of biological pattern information provided by an exemplary embodiment of the present application;
[0083] Figure 8 is a schematic diagram of an area of interest for capturing biometric pattern information provided by an exemplary embodiment of the present application;
[0084] Fig. 9 is a flow chart of a method for extracting features of biological pattern information provided by an exemplary embodiment of the present application;
[0085] Fig.10 is a flow chart of an identity recognition method based on biometric pattern information provided by an exemplary embodiment of the present application;
[0086] Fig.11 is a schematic diagram of the structure of a biological pattern feature extraction model provided by an exemplary embodiment of the present application;
[0087] Fig.12 is a flow chart of a method for training a biological pattern feature extraction model provided by an exemplary embodiment of the present application;
[0088] Fig.13 is a schematic diagram of determining feature prediction errors provided by an exemplary embodiment of the present application;
[0089] Fig.14 is a flow chart of a method for training a biological pattern feature extraction model provided by an exemplary embodiment of the present application;
[0090] Fig.15 is a structural block diagram of a feature extraction device for biometric pattern information provided by an exemplary embodiment of the present application;
[0091] Fig.16 is a structural block diagram of a feature extraction device for biometric pattern information provided by an exemplary embodiment of the present application;
[0092] Fig.17 It is a structural block diagram of a server provided by an exemplary embodiment of the present application.
[0093] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application. DETAILED DESCRIPTION
[0094] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.
[0095] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.
[0096] The terms used in the present disclosure are only for the purpose of describing specific embodiments, and are not intended to limit the present disclosure. The singular forms of "a", "said" and "the" used in the present disclosure and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items. It should be understood that although the terms first, second, etc. may be used to describe various information in the present disclosure, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present disclosure, the first parameter may also be referred to as the second parameter, and similarly, the second parameter may also be referred to as the first parameter. Depending on the context, the word "if" as used herein can be interpreted as "at the time of" or "when" or "in response to determination".
[0097] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions. For example, the biometric information involved in this application is obtained with full authorization.
[0098] Figure 1 A schematic diagram of a computer system provided by an exemplary embodiment of the present application is shown. The computer system can be implemented as a system architecture of a feature extraction method for biometric pattern information. The computer system may include: a terminal 100 and a server 200.
[0099] The terminal 100 may be an electronic device such as a mobile phone, a tablet computer, a vehicle-mounted terminal (vehicle computer), a wearable device, a PC (Personal Computer), an access control device, an unmanned vending terminal, etc. The terminal 100 may be installed with a client that runs a target application, and the target application may be a mobile payment, office attendance, or identity authentication application that supports biometric pattern information verification, or may be other applications that provide a feature extraction function for biometric pattern information, which is not limited in this application. In addition, this application does not limit the form of the target application, including but not limited to an application (Application, App) installed in the terminal 100, a mini-program, etc., and may also be in the form of a web page.
[0100] The server 200 may be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The server 200 may be a background server of the target application, used to provide background services for the client of the target application.
[0101] In the method for extracting the feature of biological pattern information provided in the embodiment of the present application, the execution subject of each step may be a computer device, which refers to an electronic device with data calculation, processing and storage capabilities. Figure 1Taking the implementation environment of the scheme shown as an example, the feature extraction method of biometric pattern information can be executed by the terminal 100 (such as the client of the target application installed and running in the terminal 100 executes the feature extraction method of the biometric pattern information), or the server 200 executes the feature extraction method of the biometric pattern information, or the terminal 100 and the server 200 interact and cooperate to execute, and this application is not limited to this. In addition, the technical solution of the present application can be combined with blockchain technology. For example, the feature extraction method of biometric pattern information disclosed in this application, some of the data involved (such as biometric pattern information, identity information corresponding to the biometric pattern information, and other data) can be stored on the blockchain. The terminal 100 and the server 200 can communicate through a network, such as a wired or wireless network.
[0102] Next, the feature extraction method of biological pattern information will be introduced through the following embodiments.
[0103] Figure 2 A flowchart of a method for extracting features of biological pattern information provided by an exemplary embodiment of the present application is shown. The method can be executed by a computer device. The method includes:
[0104] Step 510: Obtaining biometric pattern information;
[0105] The biometric pattern information is used to describe the biometric pattern of the first object; in the embodiment of the present application, the type of the biometric pattern includes but is not limited to at least one of the following: finger patterns, palm patterns, eye patterns, facial patterns, blood vessel patterns, lip patterns, and oral patterns; the present application does not make any restrictive provisions on the type of the biometric pattern;
[0106] In the embodiments of the present application, the bioprint collection method includes but is limited to at least one of the following: optical collection method, such as bioprint photos; capacitive collection method, such as using a capacitive print collection module; biological radio frequency collection method, such as using ultrasound to collect bioprints. This application does not make any restrictive provisions on the bioprint collection method.
[0107] Step 520: performing feature extraction processing on the biological texture information to obtain global texture features and local texture features;
[0108] The global texture feature is used to describe the global feature of the biological texture information of the first object, and the local texture feature is used to describe the local feature of the biological texture information of the first object; in the embodiment of the present application, the local texture feature can be obtained based on the global texture feature, or the local texture feature and the global texture feature can be obtained independently of each other; no restrictive provisions are made for this;
[0109] It should be noted that the feature extraction process of the biological pattern information can be completed by calling a biological pattern feature extraction model to complete the feature extraction of the biological pattern information; the feature extraction of the biological pattern information can also be completed by a manual annotation method; or the feature of the biological pattern information can be extracted by calculating the biological pattern information. This application does not make any restrictive provisions on this.
[0110] Exemplarily, taking the example of performing feature extraction processing on biological pattern information through a biological pattern feature extraction model: the biological pattern feature extraction model includes a feature extraction network; calling the feature extraction network to perform feature extraction processing on the biological pattern information to obtain global pattern features and local pattern features, the global pattern features are used to describe the global features of the biological pattern information of the first object, and the local pattern features are used to describe the local features of the biological pattern information of the first object.
[0111] Step 530: fusing the global texture features and the local texture features of the first object to obtain a fused texture feature of the first object;
[0112] Exemplarily, the fused texture features are used to describe the global features and local features of the biological texture information of the first object; in an embodiment of the present application, the way of fusing the global texture features and the local texture features includes but is not limited to at least one of the following: orthogonal fusion, convolution fusion, and adaptive feature fusion.
[0113] It should be noted that, similar to step 520 above, the present application does not make any restrictive provisions on the manner of performing the fusion processing; taking the fusion processing of global texture features and local texture features through a biological texture feature extraction model as an example: the biological texture feature extraction model includes a feature fusion network; the feature fusion network is called to fuse the global texture features and local texture features of the first object to obtain the fused texture features of the first object.
[0114] In summary, the method provided in this embodiment obtains local texture features by performing feature extraction processing on biometric texture information, thereby improving the attention to local detail features in biometric textures; integrating local texture features with global texture features, thereby improving the ability of texture features to describe palmprint information; and expanding the dimension of texture features to describe palmprint information from a local perspective, thereby ensuring recognition accuracy when facing biometric texture information with high similarity.
[0115] Next, the feature extraction process is introduced. Figure 3 A flowchart of a method for extracting features of biological pattern information provided by an exemplary embodiment of the present application is shown. The method can be executed by a computer device. That is, in an optional design, Figure 2 Step 520 in the illustrated embodiment may be implemented as the following two steps:
[0116] Step 522: extracting latent features from the biological texture information to obtain texture latent features of the first object;
[0117] The texture hidden layer feature is used to describe the hidden layer representation of the target palmprint information; exemplarily, the representation method of the texture hidden layer feature includes but is not limited to at least one of the following: feature vector, feature matrix, feature value;
[0118] It should be noted that, similar to step 520 above, the present application does not make any restrictive provisions on the manner of performing hidden layer feature extraction; taking the hidden layer feature extraction of biological pattern information through the biological pattern feature extraction model as an example: the biological pattern feature extraction model includes a feature extraction network; the feature extraction network includes a first residual convolution block; calling the first residual convolution block to extract the hidden layer feature of the biological pattern information, and obtaining the texture hidden layer feature of the first object, the texture hidden layer feature is used to describe the hidden layer representation of the target palmprint information;
[0119] Exemplarily, the feature extraction method for performing hidden layer feature extraction on biological pattern information is a feature extraction method based on reducing feature dimension; in the process of performing hidden layer feature extraction through the biological pattern feature extraction model, the first residual convolution block is called to realize hidden layer feature extraction.
[0120] Optionally, the first residual convolution block includes but is not limited to at least one of the following network structures: Convolutional Neural Networks (CNN), Efficient Networks (Efficient Networks), Transformer Networks (Transformer Networks), Deep residual network (Deep residual network, ResNet), and Inception-Resnet101.
[0121] Step 524: extracting the texture hidden layer features by a first feature extraction method to obtain the global texture features of the first object;
[0122] The first feature extraction method is a feature extraction method based on reducing feature dimensions; illustratively, the feature dimensions are reduced by residual convolution.
[0123] Similar to step 520 above, the present application does not make any restrictive provisions on the method of performing feature extraction; taking the extraction of hidden layer features of biological texture information through a biological texture feature extraction model as an example: the biological texture feature extraction model includes a feature extraction network; the feature extraction network includes a second residual convolution block; calling the second residual convolution block, extracting the texture hidden layer features through a first feature extraction method, and obtaining the global texture features of the first object, the first feature extraction method is a feature extraction method based on reducing the feature dimension;
[0124] Step 526: extracting the texture hidden layer features by a second feature extraction method to obtain local texture features of the first object;
[0125] The second feature extraction method is a feature extraction method based on increasing the feature receptive field. Exemplarily, the feature receptive field is increased by a dilated convolution method.
[0126] Exemplarily, the feature receptive field is used to indicate the information dimension of the texture hidden layer feature indicated by each eigenvalue in the local texture feature. Exemplarily, in the process of extracting features from the first original image using the first feature image, the feature receptive field represents the dimension of the original image pixel indicated by an image feature pixel in the first feature image.
[0127] Similar to step 520 above, the present application does not make any restrictive provisions on the method of performing feature extraction; taking the biological texture feature extraction model to extract hidden layer features of biological texture information as an example: the biological texture feature extraction model includes a feature extraction network; the feature extraction network includes a hole convolution block; calling the hole convolution block, extracting the texture hidden layer features through the second feature extraction method, and obtaining the local texture features of the first object, the second feature extraction method is a feature extraction method based on increasing the feature receptive field;
[0128] In an optional design of this embodiment, step 526 can be implemented as the following two sub-steps:
[0129] Sub-step 1: Processing the texture hidden layer features, respectively using n second feature extraction methods to obtain n local sub-features of the first object;
[0130] n is a positive integer, and the feature receptive fields of n second feature extraction methods are different;
[0131] Similar to step 520 above, the present application does not make any restrictive provisions on the method of performing feature extraction; take the hidden layer feature extraction of biological texture information through the biological texture feature extraction model as an example: the biological texture feature extraction model includes a feature extraction network; the feature extraction network includes a hole convolution block; the hole convolution block includes n hole convolution layers; call n hole convolution layers to process the texture hidden layer features, and obtain n local sub-features of the first object through n second feature extraction methods respectively, n is a positive integer, and the feature receptive fields of the n second feature extraction methods are different, and the feature receptive field is determined based on the convolution kernel scale of the hole convolution layer.
[0132] Sub-step 2: splicing n local sub-features to obtain a local texture feature of the first object;
[0133] When n is greater than 1, the local sub-features may be spliced together directly end to end, or the local sub-features may be superimposed and spliced together in a mathematical manner; this embodiment does not impose any limitation on this.
[0134] When n is equal to 1, the local sub-feature is determined as the local feature of the first object;
[0135] For example, Figure 4 A schematic diagram of a dilated convolution kernel provided by an exemplary embodiment of the present application is shown, and a schematic diagram of a dilated convolution block shows a dilated convolution kernel with a convolution kernel scale of 3, a dilated convolution kernel with a convolution kernel scale of 6, and a dilated convolution kernel with a convolution kernel scale of 9. Among them, the feature receptive field dimension of the dilated convolution kernel with a convolution kernel scale of 3 is 5 times 5, the feature receptive field dimension of the dilated convolution kernel with a convolution kernel scale of 6 is 11 times 11, and the feature receptive field dimension of the dilated convolution kernel with a convolution kernel scale of 9 is 17 times 17.
[0136] In summary, the method provided in this embodiment obtains local texture features based on a feature extraction method that increases the feature receptive field by performing feature extraction processing on biometric texture information, thereby increasing the focus on local detail features in biometric textures; integrating local texture features with global texture features, thereby improving the ability of texture features to describe palmprint information; and expanding the dimension of texture features to describe palmprint information locally, thereby ensuring recognition accuracy when facing biometric texture information with high similarity.
[0137] Next, the self-attention mechanism is introduced. Figure 5 The flowchart of a method for extracting features of biological pattern information provided by an exemplary embodiment of the present application is shown. The method can be executed by a computer device. That is, in an optional design, Figure 2 The illustrated embodiment also includes step 528:
[0138] Step 528: Perform self-attention mechanism processing on the local texture features to obtain updated local texture features;
[0139] Exemplarily, a self-attention mechanism is performed on the local texture features to enhance the weight of the local effective feature area, so as to obtain an updated local texture feature;
[0140] Exemplarily, in an optional design of this embodiment, step 528 can be implemented as the following three sub-steps, that is, the self-attention mechanism processing includes activation processing and regularization processing:
[0141] Sub-step 1: Activate the local texture feature to obtain the second activated feature;
[0142] Exemplarily, the activation function for activating the local texture features includes but is not limited to at least one of the following: Sigmoid function, Tanh function, ReLU function, Softmax function, Softplus function; Figure 6 A schematic diagram of a ReLU function and a Softplus function provided by an exemplary embodiment of the present application is shown.
[0143] Exemplarily, the local texture features are activated by ReLU function and Softplus function to obtain the second activation feature, for example:
[0144] x relu =Relu(x)=max(0,x);
[0145]
[0146] Where x represents the local texture feature, Relu(.) represents the ReLU function, and max represents the maximum value operation; x softplus represents the second activation feature, and Softplus(.) represents the Softplus function.
[0147] For example, the ReLU function can increase the nonlinear relationship between the neural network layers in the biological texture feature extraction model to avoid the appearance of a simple linear relationship. The ReLU function outputs some local texture features as 0 through the maximum value operation, resulting in the sparsity of the network and reducing the interdependence between local texture features, alleviating the occurrence of overfitting problems. The Softplus function has the same function as the ReLU function, and the Softplus function is a smoothed form of the ReLU function.
[0148] Sub-step 2: Regularize the local texture features to obtain the second regularized features;
[0149] Exemplarily, the regularization method for local texture features can be L1 parameter regularization or L2 parameter regularization; Exemplarily, the local texture features are regularized by L2 parameter regularization to obtain a second regularized feature, for example:
[0150]
[0151] Among them, x L2norm represents the second regular feature, L2norm represents L2 parameter regularization;
[0152] Exemplarily, L2 parameter regularization controls the value of the second regular feature between 0 and 1 by calculating the Euclidean distance, which effectively prevents overfitting of the second regular feature and improves the generalization ability of the model.
[0153] Sub-step 3: multiplying the second activation feature and the second regular feature to obtain an updated local texture feature of the first object;
[0154] Optionally, in an optional design of this embodiment, before sub-step 1, the method further includes: normalizing the local texture features to obtain second normalized features;
[0155] Correspondingly, sub-step 1 is implemented by performing regularization processing on the second normalized feature to obtain the second regularized feature; sub-step 2 is implemented by performing regularization processing on the second normalized feature to obtain the second regularized feature.
[0156] The local texture features are normalized to ensure that the second normalized features are distributed within a fixed interval, laying the foundation for the model to quickly obtain convergence results; exemplary, the local texture features are normalized as follows:
[0157]
[0158]
[0159]
[0160] x BN =γ.x norm +β;
[0161] Among them, r mean represents the feature average of local texture features, x represents the local texture features, x i represents the i-th eigenvalue of the local texture feature, m represents the number of eigenvalues of the local texture feature; r std Represents the characteristic standard deviation of local texture features; x normrepresents the normalized features corresponding to the local texture features, ε represents the additive variable, which is used to avoid the inability to perform gradient backpropagation due to the variance being 0; x BN represents the second normalized feature, γ and β represent the network learnable parameters, which are used to fit the distribution of the second normalized feature.
[0162] Step 530a: fusing the global texture features of the first object and the updated local texture features to obtain fused texture features of the first object;
[0163] Exemplarily, the fused texture features are used to describe the global features and local features of the biological texture information of the first object; in an embodiment of the present application, the way of fusing the global texture features and the local texture features includes but is not limited to at least one of the following: orthogonal fusion, convolution fusion, and adaptive feature fusion.
[0164] Exemplarily, the global texture features of the first object and the updated local texture features are fused in an orthogonal fusion manner to obtain the fused texture features of the first object:
[0165]
[0166] Among them, feature local-global Indicates the fusion texture feature, feature global Represents the global texture features, feature local Represents the updated local texture features.
[0167] In summary, the method provided in this embodiment obtains local texture features by performing feature extraction processing on biometric texture information, thereby improving the attention to local detail features in biometric textures, and enhancing the weight of local effective feature areas through self-attention mechanism processing; the local texture features are integrated with the global texture features to improve the description ability of the texture features for palmprint information; the dimension of the texture features from the local description of palmprint information is expanded, and the recognition accuracy is ensured when facing high-similarity biometric texture information.
[0168] Next, we will introduce the preprocessing. Figure 7 The flowchart of a method for extracting features of biological pattern information provided by an exemplary embodiment of the present application is shown. The method can be executed by a computer device. That is, in an optional design, Figure 2 The illustrated embodiment further includes step 512:
[0169] Step 512: pre-processing the biological pattern information to obtain updated biological pattern information;
[0170] Exemplarily, the preprocessing method includes but is not limited to at least one of the following:
[0171] Interpolation processing is performed on the biological pattern information; illustratively, the interpolation processing is performed on the biological pattern information to change the dimension of the biological pattern information;
[0172] Intercepting a region of interest (ROI) of the biometric pattern information; illustratively, intercepting a region of interest of the biometric pattern information is used to delineate a key area of interest;
[0173] Exemplarily, interpolation processing is performed on the biological pattern information; it is used to change the dimension of the biological pattern information and change the biological pattern information to a target dimension; for example, the biological pattern information is processed into a feature matrix with a dimension of 224 times 224 by interpolation processing on the biological pattern information;
[0174] Exemplarily, this embodiment does not make any restrictive provisions on the method of intercepting the region of interest of the biometric pattern information. The region of interest can be determined by a calculation rule or by a region of interest extraction network.
[0175] Figure 8 A schematic diagram of an area of interest for intercepting biometric pattern information provided by an exemplary embodiment of the present application is shown; taking the biometric pattern information as palm pattern image information as an example, a first key point A, a second key point B, and a third key point C are extracted from the palm pattern image information; wherein the first key point A is a gap point between the second finger and the third finger of the palm; the second key point B is a gap point between the third finger and the fourth finger of the palm; and the third feature point C is a gap point between the fourth finger and the fifth finger of the palm;
[0176] A rectangular coordinate system is established in the palm pattern image information, wherein the direction from the third key point C to the first key point A is determined as the positive direction of the x-axis, the perpendicular intersection point E with the x-axis direction is determined through the second key point B, and the direction from the vertical focus E to the second key point B is determined as the positive direction of the y-axis; extending from the perpendicular intersection point E along the negative direction of the y-axis, the palm point D is determined, and the region of interest of the palm pattern image information is determined with the palm point D as the center. The region of interest of the palm pattern image information is usually a square, but the possibility that the region of interest of the palm pattern image information is a rectangle, a circle, a sector or other shapes is not excluded. Exemplarily, the distance between the palm point D and the vertical focus E is six-fifths of the distance from the first key point A to the third key point C; the region of interest of the palm pattern image information is a square area with a side length of two-thirds of the distance from the first key point A to the third key point C centered on the palm point D.
[0177] Step 520a: performing feature extraction processing on the updated biological texture information to obtain global texture features and local texture features;
[0178] The global texture feature is used to describe the global feature of the biological texture information of the first object, and the local texture feature is used to describe the local feature of the biological texture information of the first object; in the embodiment of the present application, the local texture feature can be obtained based on the global texture feature, or the local texture feature and the global texture feature can be obtained independently of each other; no restrictive provisions are made for this;
[0179] In summary, the method provided in this embodiment obtains local texture features by performing feature extraction processing on biometric texture information, thereby increasing the attention to local detail features in biometric textures; lays the foundation for obtaining local texture features through preprocessing; integrates local texture features with global texture features, thereby improving the ability of texture features to describe palmprint information; expands the dimension of texture features from the local description of palmprint information, thereby ensuring recognition accuracy when facing biometric texture information with high similarity.
[0180] Next, we will introduce identity judgment. Fig. 9 The flowchart of a method for extracting features of biological pattern information provided by an exemplary embodiment of the present application is shown. The method can be executed by a computer device. That is, in an optional design, Figure 2 The illustrated embodiment further includes step 540:
[0181] Step 540: Compare the fused texture features of the first object with the sample texture features to obtain an identity determination result of the first object;
[0182] The sample texture feature is used to describe the biological texture information feature of the target object; the identity judgment result of the first object is used to indicate whether the first object and the target object belong to the same identity;
[0183] Exemplarily, a method of comparing the fused texture features of the first object with the sample texture features includes but is not limited to at least one of the following: calculating cosine similarity, calculating Euclidean distance, calculating covariance distance, and calculating Chebyshev distance.
[0184] Optionally, in an optional design of this embodiment, step 540 may be implemented as the following sub-steps:
[0185] Calculate the cosine similarity between the fusion texture features and the sample texture features;
[0186] Exemplarily, calculating the cosine similarity includes:
[0187]
[0188] Among them, sim(vector reg ,vector rec ) represents cosine similarity, vector recIndicates the fusion texture features, vector reg represents the sample texture feature, and |||| represents the 1-norm operation.
[0189] When the cosine similarity satisfies the identity judgment condition, the identity judgment result of the first object is obtained as the target object;
[0190] When the cosine similarity does not satisfy the identity judgment condition, the identity judgment result of the first object is obtained as not being the target object.
[0191] Exemplarily, the identity determination condition is that the cosine similarity is greater than a target threshold; wherein the target threshold may be set based on experience or obtained through a training process.
[0192] In summary, the method provided in this embodiment obtains local pattern features by performing feature extraction processing on biometric pattern information, thereby increasing the focus on local detail features in biometric patterns; completing identity judgment by comparing pattern features; and expanding the dimension of pattern features from locally describing palm print information, thereby ensuring recognition accuracy when facing biometric pattern information with high similarity.
[0193] Next, the application of the feature extraction method of biometric pattern information in the field of mobile payment is explained; Fig.10 A flowchart of an identity recognition method based on biometric information provided by an exemplary embodiment of the present application is shown. The method can be executed by a computer device. The method includes:
[0194] Step 610: Obtaining the biometric pattern information of the first object through the terminal payment device;
[0195] The biometric pattern information is used to describe the biometric pattern of the first object; in the embodiment of the present application, the type of the biometric pattern includes but is not limited to at least one of the following: finger patterns, palm patterns, eye patterns, facial patterns, blood vessel patterns, lip patterns, and oral patterns; in the present embodiment, the biometric pattern is a palm pattern as an example, that is, the biometric pattern information is introduced as palm pattern image information;
[0196] Step 620: Detect the biometric pattern information through the terminal payment system to obtain the key points of the finger joints;
[0197] Exemplarily, the key points of the finger joints are obtained to establish a rectangular coordinate system in the palm pattern image information; and the region of interest is intercepted in the palm pattern image information by establishing the rectangular coordinate system.
[0198] In this embodiment, step 620 may be performed by a terminal payment device in the terminal payment system, or may be performed by a terminal payment server in the terminal payment system, and this embodiment does not make any limitation.
[0199] Similarly, no restrictions are imposed on the execution manner of step 630 to step 670.
[0200] Step 630: The terminal payment system captures the region of interest of the biometric pattern information based on the key points of the finger joints;
[0201] The terminal payment system captures the area of interest of the biometric pattern information based on the rectangular coordinate system established based on the key points of the finger joints.
[0202] Step 642: extracting global features of the region of interest of the biometric texture information through the terminal payment system to obtain global texture features of the first object;
[0203] Exemplarily, global feature extraction of the region of interest of biological texture information adopts a feature extraction method based on reducing feature dimension; specifically, the global texture feature of the first object is obtained by calling the residual convolution block to realize global feature extraction.
[0204] Step 644: Using the terminal payment system, three feature extraction methods with different receptive fields are used to extract local features of the region of interest of the biological texture information to obtain local texture features of the first object;
[0205] Exemplarily, local feature extraction of the region of interest of biological texture information adopts a feature extraction method based on increasing the feature receptive field; specifically, the local feature extraction is realized by calling the hole convolution block to obtain the local texture feature of the first object.
[0206] Step 646: The terminal payment system processes the local texture features using a self-attention mechanism to obtain updated local texture features;
[0207] The local texture features are processed by the self-attention mechanism to enhance the weight of the local effective feature area to obtain the updated local texture features;
[0208] Exemplarily, in an optional design of this embodiment, the self-attention mechanism processing includes normalization processing, activation processing and regularization processing;
[0209] Step 650: fusing the global texture features and the updated local texture features through the terminal payment system to obtain a fused texture feature of the first object;
[0210] Exemplarily, the fused texture features are used to describe the global features and local features of the biological texture information of the first object; specifically, the global texture features and the updated local texture features are fused in an orthogonal fusion manner to obtain the fused texture features of the first object.
[0211] Step 660: Calculate the cosine similarity between the fusion texture feature and the sample texture feature through the terminal payment system;
[0212] Exemplarily, the cosine similarity indicates the similarity between the fused texture feature and the sample texture feature through the cosine angle between the vectors.
[0213] Step 670: Determine, through the terminal payment system, the sample object corresponding to the sample with the highest cosine similarity as the identity recognition result of the first object;
[0214] The cosine similarities are arranged in descending order, and the sample texture feature with the highest cosine similarity indicates that the similarity between the sample texture feature and the fusion texture feature is the highest; the sample object corresponding to the sample texture feature is determined as the identity recognition result of the first object.
[0215] In summary, the method provided in this embodiment obtains local texture features by performing feature extraction processing on biometric texture information, thereby improving the attention to local detail features in biometric textures; integrating local texture features with global texture features, thereby improving the ability of texture features to describe palmprint information; and expanding the dimension of texture features to describe palmprint information from a local perspective, thereby ensuring recognition accuracy when facing biometric texture information with high similarity.
[0216] Next, the biological pattern feature extraction model is explained; Figure 2 In the illustrated embodiment, it is explained that a feature extraction method of biological pattern information can be implemented by a biological pattern feature extraction model; Fig.11 A schematic diagram of the structure of a biological pattern feature extraction model provided by an exemplary embodiment of the present application is shown;
[0217] The biological pattern feature extraction model includes: a feature extraction network 330, a self-attention network 350 and a feature fusion network 360;
[0218] The biological texture feature 310 is processed by intercepting the interest region 322 to obtain the texture information 310a of the interest region, and the texture information 310a of the interest region is interpolated 324 to obtain the interpolated texture information 310b; illustratively, the interpolated texture information 310b is a feature matrix with a dimension of 224 times 224;
[0219] The feature extraction network 330 includes: a first residual convolution block 332, a second residual convolution block 334 and a dilated convolution block 336;
[0220] The first residual convolution block 332 is called to process the interpolated texture information 310b to obtain texture hidden layer features. Exemplarily, the texture hidden layer features are feature matrices with a dimension of 28 times 28; wherein the first residual convolution block 332 includes three residual convolution layers, and the dimension of the feature matrix output by the residual convolution layer is one-fourth of the dimension of the input feature matrix, that is, the first residual convolution layer is called to process the interpolated texture information 310b, and the dimension of the output feature matrix is 112 times 112; the second residual convolution layer is called to process the above feature matrix, and the dimension of the output feature matrix is 56 times 56; the third residual convolution layer is called to process the above feature matrix, and the dimension of the output feature matrix is 28 times 28;
[0221] The second residual convolution block 332 is called to process the texture hidden layer features to obtain a global texture feature 342; illustratively, the global texture feature 342 is a feature matrix with a dimension of 14 by 14.
[0222] The dilated convolution block 336 includes: a first dilated convolution layer 336a, a second dilated convolution layer 336b and a third dilated convolution layer 336c;
[0223] The first atrous convolution layer 336a is called to process the texture hidden layer features to obtain the first local sub-features, and the atrous convolution kernel scale of the first atrous convolution layer 336a is 3; the second atrous convolution layer 336b is called to process the texture hidden layer features to obtain the second local sub-features, and the atrous convolution kernel scale of the second atrous convolution layer 336b is 6; the third atrous convolution layer 336c is called to process the texture hidden layer features to obtain the third local sub-features, and the atrous convolution kernel scale of the third atrous convolution layer 336c is 9.
[0224] The first local sub-feature, the second local sub-feature and the third local sub-feature are spliced to obtain an initial local texture feature 344;
[0225] The self-attention network 350 includes: a normalization layer 352, an activation layer 354, and a regularization layer 356;
[0226] Calling the normalization layer to process the initial local texture features 344, ensuring that the updated local texture features 344a are distributed within a fixed interval, laying a foundation for the model to quickly obtain convergence results;
[0227] The activation layer 354 and the regularization layer 356 are called to perform activation processing and regularization processing on the normalized local texture features respectively; the processing results are multiplied to obtain the updated local texture features 344a;
[0228] Multiplying the updated local texture feature 344a with the initial local texture feature 344 to obtain a local texture feature 344c;
[0229] The feature fusion network 360 is called to fuse the global texture feature 342 and the local texture feature 344c of the first object to obtain a fused texture feature 346 of the first object.
[0230] Calculating the cosine similarity 352 between the fused texture feature 346 and the sample texture feature 348;
[0231] The identity determination result 370 of the first target is determined according to the cosine similarity 352 .
[0232] That is, when the cosine similarity 352 satisfies the identity judgment condition, the identity judgment result 370 of the first object is the target object;
[0233] When the cosine similarity 352 does not satisfy the identity determination condition, the identity determination result 370 of the first object is obtained to be not the target object.
[0234] Next, the training of the biological pattern feature extraction model is introduced.
[0235] Fig.12 A flowchart of a method for training a biological pattern feature extraction model provided by an exemplary embodiment of the present application is shown. The method can be executed by a computer device. The method includes:
[0236] Step 710: Obtain sample texture information and sample texture features;
[0237] The sample texture information is used to describe the biological texture of the sample object, and the sample texture feature is the biological texture feature corresponding to the sample texture information of the sample object;
[0238] In the embodiments of the present application, the types of biological patterns include but are not limited to at least one of the following: finger patterns, palm patterns, eye patterns, facial patterns, blood vessel patterns, lip patterns, and oral patterns; the present application does not make any restrictive provisions on the types of biological patterns;
[0239] Step 720: calling the initial biological texture feature extraction model to process the sample texture information and output the predicted fused texture features of the sample object;
[0240] Exemplarily, the predicted fused texture features are used to describe the global features and local features of the biological texture information of the sample object;
[0241] Step 730: Compare the predicted fusion texture features with the sample texture features to obtain feature prediction errors;
[0242] The sample texture feature is a biological texture feature corresponding to the sample texture information of the sample object;
[0243] Exemplarily, the feature prediction error is a normalized exponential function (Softmax) loss function between the predicted fusion texture feature and the sample texture feature;
[0244] Specifically, the feature prediction error is obtained, including:
[0245]
[0246]
[0247]
[0248] θ yj =arccos(cos(θ yj ));
[0249] cos(θ j )=cos(θ yj +m);
[0250]
[0251] Among them, x i represents the predicted fusion texture feature, i represents the dimension of the predicted fusion texture feature, L2norm represents L2 parameter regularization, and w j represents the sample texture feature, j represents the dimension of the sample texture feature, the dimension of w is n times d, n represents the total number of sample texture features; arccos represents the arccosine operation; m represents the custom additive angle, for example, m = 0.5; s represents the feature magnification, for example, s = 48, Loss j represents the feature prediction error, and Softmax(.) represents the softmax function. For example, Fig.13 A schematic diagram of determining feature prediction errors provided by an exemplary embodiment of the present application is shown.
[0252] Step 740: Use the feature prediction error to perform backward error propagation training on the initial biometric pattern feature extraction model to obtain the biometric pattern feature extraction model.
[0253] The purpose of backward error propagation training on the initial biological texture feature extraction model is to minimize the feature prediction error between the predicted fusion texture features and the sample texture features. The biological texture feature extraction model is obtained by training the initial biological texture feature extraction model.
[0254] In summary, the method provided in this embodiment, by performing backward error propagation training on the initial bio-texture feature extraction model, lays a foundation for fusing local texture features with global texture features and improving the description ability of texture features for palmprint information; taking the bio-texture information as palm texture image information as an example, the bio-texture feature extraction model has achieved good training results. Table 1 shows the results of identity recognition for a public set of clear and blurred palm texture image information including forty pairs of twins; the left hand / right hand of the same pair of twins is taken as a sample pair, for 3600 sample pairs; the bio-texture feature extraction model did not make any error when facing clear palm texture image information, and only made one error when facing blurred palm texture image information; compared with 37 errors and 46 errors that occurred in traditional models in the related art, significant progress has been made.
[0255] Table 1
[0256] Method / Number of incorrect sample pairs identified Clear palm pattern image information Blurred palm pattern image information Traditional Model 37 46 Bio-texture feature extraction model 0 1
[0257] Fig.14 A flowchart of a method for training a biological pattern feature extraction model provided by an exemplary embodiment of the present application is shown. The method can be executed by a computer device. The method includes:
[0258] Step 750: Obtain sample texture information and classification labels of the sample texture information;
[0259] The sample texture information is used to describe the biological texture of the sample object, and the classification label is used to indicate the sample object;
[0260] In the embodiments of the present application, the types of biological patterns include but are not limited to at least one of the following: finger patterns, palm patterns, eye patterns, facial patterns, blood vessel patterns, lip patterns, and oral patterns; the present application does not make any restrictive provisions on the types of biological patterns;
[0261] Step 760: calling the initial biological texture feature extraction model to process the sample texture information and output the predicted fused texture features of the sample object;
[0262] Exemplarily, the predicted fused texture features are used to describe the global features and local features of the biological texture information of the sample object;
[0263] Step 770: performing label classification processing on the predicted fused texture features to obtain a predicted label of the sample object;
[0264] Exemplarily, in this embodiment, the label classification process may be implemented by calling a label prediction model to implement the label classification process of the predicted fused texture features; the label classification process may also be completed by a manual annotation method; or the predicted label of the sample object may be obtained by calculating the predicted fused texture features; this embodiment does not make any restrictive provisions on the implementation method of the label classification process;
[0265] Step 780: Compare the predicted label with the classification label to obtain the label prediction error;
[0266] The classification label is used to indicate the sample object; illustratively, the label prediction error includes at least one of the following loss functions between the predicted label and the classification label: 0-1 loss function (Zero-One Loss), absolute value loss function, logarithmic loss function.
[0267] Step 790: Use the label prediction error to perform backward error propagation training on the initial biometric pattern feature extraction model to obtain the biometric pattern feature extraction model.
[0268] The purpose of performing backward error propagation training on the initial bioprint feature extraction model is to minimize the feature prediction error between the predicted label and the classification label. The bioprint feature extraction model is obtained by training the initial bioprint feature extraction model.
[0269] In summary, the method provided in this embodiment, by performing backward error propagation training on the initial biometric pattern feature extraction model, lays a foundation for fusing local pattern features with global pattern features and improving the ability of pattern features to describe palmprint information.
[0270] Those skilled in the art can understand that the above embodiments can be implemented independently, or the above embodiments can be freely combined to form new embodiments to implement the feature extraction method of biometric pattern information of the present application.
[0271] Fig.15 A block diagram of a feature extraction device for biometric pattern information provided by an exemplary embodiment of the present application is shown. The device comprises:
[0272] An acquisition module 810 is used to acquire biological pattern information, where the biological pattern information is used to describe the biological pattern of the first object;
[0273] An extraction module 820, configured to perform feature extraction processing on the biological texture information to obtain global texture features and local texture features, wherein the global texture features are used to describe global features of the biological texture information of the first object, and the local texture features are used to describe local features of the biological texture information of the first object;
[0274] The fusion module 830 is used to fuse the global texture features and the local texture features of the first object to obtain fused texture features of the first object.
[0275] In an optional design of this embodiment, the extraction module 820 is further used for:
[0276] Performing hidden layer feature extraction on the biometric pattern information to obtain texture hidden layer features of the first object, wherein the texture hidden layer features are used to describe the hidden layer representation of the target palmprint information;
[0277] Extracting the texture hidden layer features by a first feature extraction method to obtain the global texture features of the first object, wherein the first feature extraction method is a feature extraction method based on reducing feature dimensions;
[0278] The texture hidden layer features are extracted by a second feature extraction method to obtain the local texture features of the first object, wherein the second feature extraction method is a feature extraction method based on increasing the feature receptive field.
[0279] In an optional design of this embodiment, the extraction module 820 is further used for:
[0280] Processing the texture hidden layer features, respectively using n second feature extraction methods to obtain n local sub-features of the first object, where n is a positive integer, and the feature receptive fields of the n second feature extraction methods are different;
[0281] The n local sub-features are spliced to obtain the local texture feature of the first object.
[0282] In an optional design of this embodiment, the device further includes: a processing module 840, configured to perform a self-attention mechanism processing on the local texture feature to obtain an updated local texture feature;
[0283] The fusion module 830 is further used to fuse the global texture features and the updated local texture features of the first object to obtain fused texture features of the first object.
[0284] In an optional design of this embodiment, the self-attention mechanism processing includes activation processing and regularization processing;
[0285] The processing module 840 is further configured to:
[0286] Performing the activation process on the local texture feature to obtain a second activation feature;
[0287] Performing the regularization process on the local texture feature to obtain a second regularized feature;
[0288] The second activation feature and the second regular feature are multiplied to obtain the updated local texture feature of the first object.
[0289] In an optional design of this embodiment, the device further includes: a processing module 840, which is used to pre-process the biometric information to obtain updated biometric information; the pre-processing method includes but is not limited to at least one of the following: interpolation processing of the biometric information; intercepting the interest area of the biometric information;
[0290] The extraction module 820 is further used to perform feature extraction processing on the updated biological texture information to obtain global texture features and local texture features.
[0291] In an optional design of this embodiment, the device further includes:
[0292] The judgment module 850 is used to compare the fused texture features of the first object with the sample texture features to obtain the identity judgment result of the first object. The sample texture features are used to describe the biological texture information features of the target object.
[0293] In an optional design of this embodiment, the judging module 850 is further configured to:
[0294] Calculating the cosine similarity between the fused texture feature and the sample texture feature;
[0295] When the cosine similarity satisfies an identity judgment condition, obtaining the identity judgment result of the first object is the target object;
[0296] When the cosine similarity does not satisfy the identity judgment condition, the identity judgment result obtained for the first object is not the target object.
[0297] Fig.16 A block diagram of a feature extraction device for biometric information provided by an exemplary embodiment of the present application is shown. The device is executed by a biometric feature extraction model, and the biometric feature extraction model includes a feature extraction network and a feature fusion network; the device includes:
[0298] An acquisition module 860, configured to acquire biological pattern information, wherein the biological pattern information is used to describe the biological pattern of the first object;
[0299] An extraction module 870, configured to call the feature extraction network to perform feature extraction processing on the biological texture information to obtain global texture features and local texture features, wherein the global texture features are used to describe the global features of the biological texture information of the first object, and the local texture features are used to describe the local features of the biological texture information of the first object;
[0300] The fusion module 880 is used to call the feature fusion network to fuse the global texture features and the local texture features of the first object to obtain the fused texture features of the first object.
[0301] In an optional design of this embodiment, the feature extraction network includes: a first residual convolution block, a second residual convolution block and a hole convolution block; the extraction module 870 is further used to:
[0302] Calling the first residual convolution block to perform hidden layer feature extraction on the biometric pattern information to obtain texture hidden layer features of the first object, wherein the texture hidden layer features are used to describe the hidden layer representation of the target palmprint information;
[0303] Calling the second residual convolution block to extract the texture hidden layer features by a first feature extraction method to obtain the global texture features of the first object, wherein the first feature extraction method is a feature extraction method based on reducing feature dimensions;
[0304] The atrous convolution block is called to extract the texture hidden layer features through a second feature extraction method to obtain the local texture features of the first object, wherein the second feature extraction method is a feature extraction method based on increasing the feature receptive field.
[0305] In an optional design of this embodiment, the atrous convolution block includes n atrous convolution layers; the extraction module 870 is further used to: call the n atrous convolution layers to process the texture hidden layer features, and obtain n local sub-features of the first object through n second feature extraction methods respectively, where n is a positive integer, and the feature receptive fields of the n second feature extraction methods are different, and the feature receptive fields are determined based on the convolution kernel scale of the atrous convolution layer;
[0306] The n local sub-features are spliced to obtain the local texture feature of the first object.
[0307] In an optional design of this embodiment, the biological texture feature extraction model is trained in the following manner: obtaining sample texture information and sample texture features, wherein the sample texture information is used to describe the biological texture of the sample object, and the sample texture features are biological texture features corresponding to the sample texture information of the sample object;
[0308] Calling the initial biological texture feature extraction model to process the sample texture information and output the predicted fused texture features of the sample object;
[0309] Comparing the predicted fusion texture features with the sample texture features to obtain a feature prediction error;
[0310] The feature prediction error is used to perform backward error propagation training on the initial biometric pattern feature extraction model to obtain the biometric pattern feature extraction model.
[0311] In an optional design of this embodiment, the biological pattern feature extraction model is trained by: obtaining sample pattern information and a classification label of the sample pattern information, the sample pattern information is used to describe the biological pattern of the sample object, and the classification label is used to indicate the sample object;
[0312] Calling the initial biological texture feature extraction model to process the sample texture information and output the predicted fused texture features of the sample object;
[0313] Performing label classification processing on the predicted fused texture features to obtain a predicted label of the sample object;
[0314] Comparing the predicted label with the classification label to obtain a label prediction error;
[0315] The label prediction error is used to perform backward error propagation training on the initial biometric pattern feature extraction model to obtain the biometric pattern feature extraction model.
[0316] One point that needs to be explained is that the device provided in the above embodiment only uses the division of the above-mentioned functional modules as an example to implement its functions. In actual applications, the above-mentioned functions can be assigned to different functional modules according to actual needs, that is, the content structure of the device can be divided into different functional modules to complete all or part of the functions described above.
[0317] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method; the technical effects achieved by each module performing operations are the same as those in the embodiment of the method, and will not be elaborated here.
[0318] An embodiment of the present application also provides a computer device, which includes: a processor and a memory, wherein a computer program is stored in the memory; the processor is used to execute the computer program in the memory to implement the feature extraction method of biological pattern information provided by the above-mentioned method embodiments.
[0319] Optionally, the computer device is a server. Fig.17 It is a structural block diagram of a server provided by an exemplary embodiment of the present application.
[0320] Typically, the server 2300 includes: a processor 2301 and a memory 2302 .
[0321] The processor 2301 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 2301 may be implemented in at least one hardware form of digital signal processing (DSP), field programmable gate array (FPGA), and programmable logic array (PLA). The processor 2301 may also include a main processor and a coprocessor, the main processor is a processor for processing data in an awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 2301 may be integrated with a graphics processing unit (GPU), and the GPU is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 2301 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning. The memory 2302 may include one or more computer-readable storage media, which may be non-transitory. The memory 2302 may also include a high-speed random access memory and a non-volatile memory, such as one or more disk storage devices and flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 2302 is used to store at least one instruction, which is executed by the processor 2301 to implement the feature extraction method of the biometric pattern information provided in the method embodiment of the present application.
[0322] In some embodiments, the server 2300 may also optionally include: an input interface 2303 and an output interface 2304. The processor 2301, the memory 2302, the input interface 2303, and the output interface 2304 may be connected via a bus or a signal line. Each peripheral device may be connected to the input interface 2303 and the output interface 2304 via a bus, a signal line, or a circuit board. The input interface 2303 and the output interface 2304 may be used to connect at least one peripheral device related to input / output (I / O) to the processor 2301 and the memory 2302. In some embodiments, the processor 2301, the memory 2302, the input interface 2303, and the output interface 2304 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 2301, the memory 2302, the input interface 2303, and the output interface 2304 may be implemented on a separate chip or circuit board, which is not limited in the embodiments of the present application.
[0323] Those skilled in the art will appreciate that the structure shown above does not constitute a limitation on the server 2300 , and may include more or fewer components than shown in the figure, or combine certain components, or adopt a different component arrangement.
[0324] In an exemplary embodiment, a chip is also provided, which includes a programmable logic circuit and / or program instructions. When the chip is run on a computer device, it is used to implement the feature extraction method of the biometric pattern information described in the above aspects.
[0325] In an exemplary embodiment, a computer program product or a computer program is also provided, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor reads and executes the computer instructions from the computer-readable storage medium to implement the feature extraction method of the biometric pattern information provided by the above-mentioned method embodiments.
[0326] In an exemplary embodiment, a computer-readable storage medium is also provided, in which a computer program is stored. The computer program is loaded and executed by a processor to implement the feature extraction method of biological pattern information provided by the above-mentioned method embodiments.
[0327] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.
[0328] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the embodiments of the present application can be implemented with hardware, software, firmware or any combination thereof. When implemented using software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein the communication medium includes any medium that is convenient for transmitting a computer program from one place to another. The storage medium can be any available medium that a general or special-purpose computer can access. The above is only an optional embodiment of the present application, and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.
Claims
1. A feature extraction method for biological texture information, It is characterized in that The method is performed by a biological pattern feature extraction model, and the method comprises: Acquire biometric pattern information, where the biometric pattern information is used to describe the biometric pattern of the first object; the biometric pattern information includes palm pattern image information; Detecting the biometric pattern information to obtain the key points of the finger gaps; the key points of the finger gaps include a first key point, a second key point and a third key point, the first key point is a gap point between the second finger and the third finger of the palm, the second key point is a gap point between the third finger and the fourth finger of the palm, and the third key point is a gap point between the fourth finger and the fifth finger of the palm; Determine the direction from the third key point to the first key point as the positive direction of the x-axis, determine the intersection of the vertical line with the x-axis direction through the second key point, determine the direction from the intersection of the vertical line to the second key point as the positive direction of the y-axis, extend from the intersection of the vertical line along the negative direction of the y-axis to determine the palm point, determine the region of interest of the bio-texture information with the palm point as the center, and obtain the texture information of the region of interest; when the region of interest is a square area, the distance between the palm point and the intersection of the vertical line is six-fifths of the distance from the first key point to the third key point, and the side length of the square area is three-two times the distance from the first key point to the third key point; Performing interpolation processing on the texture information of the region of interest to obtain interpolated texture information; Based on a feature extraction method that reduces feature dimensions, the interpolated texture information is processed to obtain texture hidden layer features, and the texture hidden layer features are used to describe the hidden layer representation of the target palmprint information; Extracting the texture hidden layer features by a first feature extraction method to obtain a global texture feature of the first object, wherein the global texture feature is used to describe the global feature of the biological texture information of the first object, wherein the first feature extraction method is a feature extraction method based on reducing feature dimension; Extracting the texture hidden layer features by a second feature extraction method to obtain local texture features of the first object, wherein the local texture features are used to describe local features of the biological texture information of the first object, wherein the second feature extraction method is a feature extraction method based on increasing a feature receptive field; fusing the global texture features and the local texture features of the first object to obtain fused texture features of the first object; The biological pattern feature extraction model is trained in the following way: Acquire sample texture information and sample texture features, wherein the sample texture information is used to describe the biological texture of the sample object, and the sample texture features are biological texture features corresponding to the sample texture information of the sample object; Calling the initial biological texture feature extraction model to process the sample texture information and output the predicted fused texture features of the sample object; Comparing the predicted fusion texture feature with the sample texture feature to obtain a feature prediction error; The feature prediction error is used to perform backward error propagation training on the initial biometric pattern feature extraction model to obtain the biometric pattern feature extraction model.
2. The method according to claim 1, It is characterized in that The extracting the texture hidden layer features by the second feature extraction method to obtain the local texture features of the first object includes: Processing the texture hidden layer features, respectively obtaining n local sub-features of the first object through n second feature extraction methods, where n is a positive integer, and the feature receptive fields of the n second feature extraction methods are different; The n local sub-features are spliced to obtain the local texture feature of the first object.
3. The method according to any one of claims 1 to 2, It is characterized in that The method further comprises: Performing a self-attention mechanism on the local texture features to obtain updated local texture features; The fusing the global texture feature and the local texture feature of the first object to obtain a fused texture feature of the first object includes: The global texture features and the updated local texture features of the first object are fused to obtain fused texture features of the first object.
4. The method according to claim 3, It is characterized in that The self-attention mechanism processing includes activation processing and regularization processing; The performing of a self-attention mechanism on the local texture features to obtain updated local texture features includes: Performing the activation process on the local texture feature to obtain a second activation feature; Performing the regularization process on the local texture feature to obtain a second regularized feature; The second activation feature and the second regular feature are multiplied to obtain the updated local texture feature of the first object.
5. The method according to any one of claims 1 to 2, It is characterized in that The method further comprises: The fused texture features of the first object are compared with the sample texture features to obtain an identity judgment result of the first object, and the sample texture features are used to describe the biological texture information features of the target object.
6. The method according to claim 5, It is characterized in that The step of comparing the fused texture features of the first object with the sample texture features to obtain an identity determination result of the first object includes: Calculating the cosine similarity between the fused texture feature and the sample texture feature; When the cosine similarity satisfies an identity judgment condition, obtaining the identity judgment result of the first object is the target object; When the cosine similarity does not satisfy the identity judgment condition, the identity judgment result obtained for the first object is not the target object.
7. A feature extraction method for biological texture information, It is characterized in that The method is performed by a biological pattern feature extraction model, which includes a feature extraction network and a feature fusion network; The method comprises: Acquire biometric pattern information, where the biometric pattern information is used to describe the biometric pattern of the first object; the biometric pattern information includes palm pattern image information; Detecting the biometric pattern information to obtain the key points of the finger gaps; the key points of the finger gaps include a first key point, a second key point and a third key point, the first key point is a gap point between the second finger and the third finger of the palm, the second key point is a gap point between the third finger and the fourth finger of the palm, and the third key point is a gap point between the fourth finger and the fifth finger of the palm; Determine the direction from the third key point to the first key point as the positive direction of the x-axis, determine the intersection of the vertical line with the x-axis direction through the second key point, determine the direction from the intersection of the vertical line to the second key point as the positive direction of the y-axis, extend from the intersection of the vertical line along the negative direction of the y-axis to determine the palm point, determine the region of interest of the bio-texture information with the palm point as the center, and obtain the texture information of the region of interest; when the region of interest is a square area, the distance between the palm point and the intersection of the vertical line is six-fifths of the distance from the first key point to the third key point, and the side length of the square area is three-two times the distance from the first key point to the third key point; Performing interpolation processing on the texture information of the region of interest to obtain interpolated texture information; Calling the feature extraction network, processing the interpolated texture information based on a feature extraction method that reduces feature dimensions, and obtaining texture hidden layer features, wherein the texture hidden layer features are used to describe the hidden layer representation of the target palmprint information; extracting the texture hidden layer features by a first feature extraction method to obtain global texture features of the first object, wherein the global texture features are used to describe the global features of the biological texture information of the first object, wherein the first feature extraction method is a feature extraction method based on reducing feature dimensions; extracting the texture hidden layer features by a second feature extraction method to obtain local texture features of the first object, wherein the local texture features are used to describe the local features of the biological texture information of the first object, wherein the second feature extraction method is a feature extraction method based on increasing feature receptive field; Calling the feature fusion network to fuse the global texture features and the local texture features of the first object to obtain fused texture features of the first object; The biological pattern feature extraction model is trained in the following way: Acquire sample texture information and sample texture features, wherein the sample texture information is used to describe the biological texture of the sample object, and the sample texture features are biological texture features corresponding to the sample texture information of the sample object; Calling the initial biological texture feature extraction model to process the sample texture information and output the predicted fused texture features of the sample object; Comparing the predicted fusion texture feature with the sample texture feature to obtain a feature prediction error; The feature prediction error is used to perform backward error propagation training on the initial biometric pattern feature extraction model to obtain the biometric pattern feature extraction model.
8. The method according to claim 7, It is characterized in that The feature extraction network includes: a first residual convolution block, a second residual convolution block and a hole convolution block; The method further comprises: Calling the first residual convolution block to extract hidden layer features of the biological texture information to obtain texture hidden layer features of the first object; Calling the second residual convolution block to extract the texture hidden layer features by using the first feature extraction method to obtain the global texture features of the first object; The atrous convolution block is called to extract the texture hidden layer features by using the second feature extraction method to obtain the local texture features of the first object.
9. The method according to claim 8, It is characterized in that The dilated convolution block includes n dilated convolution layers; The calling of the atrous convolution block to extract the texture hidden layer features by the second feature extraction method to obtain the local texture features of the first object includes: Calling the n atrous convolution layers to process the texture hidden layer features, respectively obtaining n local sub-features of the first object through n second feature extraction methods, where n is a positive integer, and the n second feature extraction methods have different feature receptive fields, which are determined based on the convolution kernel scale of the atrous convolution layer; The n local sub-features are spliced to obtain the local texture feature of the first object.
10. A feature extraction device for biological texture information, It is characterized in that The device includes a biological pattern feature extraction model, and the device includes: An acquisition module, used for acquiring biometric pattern information, wherein the biometric pattern information is used for describing the biometric pattern of the first object; the biometric pattern information includes palm pattern image information; The processing module is used to detect the biometric pattern information and obtain the key points of the finger joints; the key points of the finger joints include a first key point, a second key point and a third key point, the first key point is a gap point between the second finger and the third finger of the palm, the second key point is a gap point between the third finger and the fourth finger of the palm, and the third key point is a gap point between the fourth finger and the fifth finger of the palm; the direction from the third key point to the first key point is determined as the positive direction of the x-axis, the intersection of the vertical line with the x-axis direction is determined through the second key point, and the intersection of the vertical line to the second key point is determined as the positive direction of the x-axis. The direction of the key point is determined as the positive direction of the y-axis, and the palm point is extended from the intersection of the vertical lines along the negative direction of the y-axis to determine the palm point, and the interest area of the biological texture information is determined with the palm point as the center to obtain the texture information of the interest area; when the interest area is a square area, the distance between the palm point and the intersection of the vertical lines is six-fifths of the distance from the first key point to the third key point, and the side length of the square area is three-two times the distance from the first key point to the third key point; the texture information of the interest area is interpolated to obtain interpolated texture information; An extraction module is used to process the interpolated texture information based on a feature extraction method that reduces feature dimensions to obtain texture hidden layer features, wherein the texture hidden layer features are used to describe the hidden layer representation of the target palmprint information; extract the texture hidden layer features by a first feature extraction method to obtain global texture features of the first object, wherein the global texture features are used to describe the global features of the biological texture information of the first object, wherein the first feature extraction method is a feature extraction method based on reducing feature dimensions; extract the texture hidden layer features by a second feature extraction method to obtain local texture features of the first object, wherein the local texture features are used to describe the local features of the biological texture information of the first object, wherein the second feature extraction method is a feature extraction method based on increasing feature receptive field; A fusion module, used for fusing the global texture features and the local texture features of the first object to obtain a fused texture feature of the first object; The biological pattern feature extraction model is trained in the following way: Acquire sample texture information and sample texture features, wherein the sample texture information is used to describe the biological texture of the sample object, and the sample texture features are biological texture features corresponding to the sample texture information of the sample object; Calling the initial biological texture feature extraction model to process the sample texture information and output the predicted fused texture features of the sample object; Comparing the predicted fusion texture feature with the sample texture feature to obtain a feature prediction error; The feature prediction error is used to perform backward error propagation training on the initial biometric pattern feature extraction model to obtain the biometric pattern feature extraction model.
11. The device according to claim 10, It is characterized in that The extraction module is further used for: Processing the texture hidden layer features, respectively obtaining n local sub-features of the first object through n second feature extraction methods, where n is a positive integer, and the feature receptive fields of the n second feature extraction methods are different; The n local sub-features are spliced to obtain the local texture feature of the first object.
12. The device according to any one of claims 10 to 11, It is characterized in that The processing module is further used to perform self-attention mechanism processing on the local texture features to obtain updated local texture features; The fusion module is further used to fuse the global texture features and the updated local texture features of the first object to obtain fused texture features of the first object.
13. The device according to claim 12, It is characterized in that The self-attention mechanism processing includes activation processing and regularization processing; The processing module is further used for: Performing the activation process on the local texture feature to obtain a second activation feature; Performing the regularization process on the local texture feature to obtain a second regularized feature; The second activation feature and the second regular feature are multiplied to obtain the updated local texture feature of the first object.
14. The device according to any one of claims 10 to 11, It is characterized in that The device also includes: The judgment module is used to compare the fused texture features of the first object with the sample texture features to obtain the identity judgment result of the first object, and the sample texture features are used to describe the biological texture information features of the target object.
15. The device according to claim 14, It is characterized in that The judging module is further used for: Calculating the cosine similarity between the fused texture feature and the sample texture feature; When the cosine similarity satisfies an identity judgment condition, obtaining the identity judgment result of the first object is the target object; When the cosine similarity does not satisfy the identity judgment condition, the identity judgment result obtained for the first object is not the target object.
16. A feature extraction device for biological texture information, It is characterized in that The device includes a biological pattern feature extraction model, and the biological pattern feature extraction model includes a feature extraction network and a feature fusion network; The device comprises: An acquisition module, used for acquiring biometric pattern information, wherein the biometric pattern information is used for describing the biometric pattern of the first object; the biometric pattern information includes palm pattern image information; The processing module is used to detect the biometric pattern information and obtain the key points of the finger joints; the key points of the finger joints include a first key point, a second key point and a third key point, the first key point is a gap point between the second finger and the third finger of the palm, the second key point is a gap point between the third finger and the fourth finger of the palm, and the third key point is a gap point between the fourth finger and the fifth finger of the palm; the direction from the third key point to the first key point is determined as the positive direction of the x-axis, the intersection of the vertical line with the x-axis direction is determined through the second key point, and the intersection of the vertical line to the second key point is determined as the positive direction of the x-axis. The direction of the key point is determined as the positive direction of the y-axis, and the palm point is extended from the intersection of the vertical lines along the negative direction of the y-axis to determine the palm point, and the interest area of the biological texture information is determined with the palm point as the center to obtain the texture information of the interest area; when the interest area is a square area, the distance between the palm point and the intersection of the vertical lines is six-fifths of the distance from the first key point to the third key point, and the side length of the square area is three-two times the distance from the first key point to the third key point; the texture information of the interest area is interpolated to obtain interpolated texture information; An extraction module is used to call the feature extraction network, process the interpolated texture information based on a feature extraction method that reduces feature dimensions, and obtain texture hidden layer features, wherein the texture hidden layer features are used to describe the hidden layer representation of the target palmprint information; extract the texture hidden layer features by a first feature extraction method to obtain global texture features of the first object, wherein the global texture features are used to describe the global features of the biological texture information of the first object, wherein the first feature extraction method is a feature extraction method based on reducing feature dimensions; extract the texture hidden layer features by a second feature extraction method to obtain local texture features of the first object, wherein the local texture features are used to describe the local features of the biological texture information of the first object, wherein the second feature extraction method is a feature extraction method based on increasing feature receptive field; A fusion module, used for invoking the feature fusion network to fuse the global texture feature and the local texture feature of the first object to obtain a fused texture feature of the first object; The biological pattern feature extraction model is trained in the following way: Acquire sample texture information and sample texture features, wherein the sample texture information is used to describe the biological texture of the sample object, and the sample texture features are biological texture features corresponding to the sample texture information of the sample object; Calling the initial biological texture feature extraction model to process the sample texture information and output the predicted fused texture features of the sample object; Comparing the predicted fusion texture feature with the sample texture feature to obtain a feature prediction error; The feature prediction error is used to perform backward error propagation training on the initial biometric pattern feature extraction model to obtain the biometric pattern feature extraction model.
17. The device according to claim 16, It is characterized in that The feature extraction network includes: a first residual convolution block, a second residual convolution block and a hole convolution block; The extraction module is further used for: Calling the first residual convolution block to extract hidden layer features of the biological texture information to obtain texture hidden layer features of the first object; Calling the second residual convolution block to extract the texture hidden layer features by using the first feature extraction method to obtain the global texture features of the first object; The atrous convolution block is called to extract the texture hidden layer features by using the second feature extraction method to obtain the local texture features of the first object.
18. The device according to claim 17, It is characterized in that The dilated convolution block includes n dilated convolution layers; The extraction module is further used for: Calling the n atrous convolution layers to process the texture hidden layer features, respectively obtaining n local sub-features of the first object through n second feature extraction methods, where n is a positive integer, and the n second feature extraction methods have different feature receptive fields, which are determined based on the convolution kernel scale of the atrous convolution layer; The n local sub-features are spliced to obtain the local texture feature of the first object.
19. A computer device, It is characterized in that The computer device comprises: a processor and a memory, wherein the memory stores at least one program; the processor is used to execute the at least one program in the memory to implement the feature extraction method of biological pattern information as described in any one of claims 1 to 9.
20. A computer-readable storage medium, It is characterized in that The readable storage medium stores executable instructions, and the executable instructions are loaded and executed by the processor to implement the feature extraction method of biological pattern information as described in any one of claims 1 to 9.
21. A computer program product, It is characterized in that The computer program product includes computer instructions, which are stored in a computer-readable storage medium. The processor reads and executes the computer instructions from the computer-readable storage medium to implement the feature extraction method of biological pattern information as described in any one of claims 1 to 9.
