Identity recognition method, biometric pattern information feature extraction method, identity recognition device, biometric pattern information feature extraction device, computer device, and computer program

Abstract

The present application discloses an identity recognition method, an apparatus, a device, and a storage medium, which belong to the technical field of identity recognition. The method includes the steps of acquiring biometric pattern information, the biometric pattern information being for describing a biometric pattern of a first object, performing a feature extraction process on the biometric pattern information to acquire a global pattern feature and a local pattern feature, performing a fusion process on the global pattern feature and the local pattern feature to acquire a fused pattern feature of the first object, and performing identity recognition on the first object based on the fused pattern feature of the first object.

Description

[Technical Field]

[0001] This application claims priority from a Chinese patent application filed with the China Patent Office on February 9, 2022, bearing application number 202210122885.7 and entitled "Method, device, equipment, and storage medium for extracting features of biometric fingerprint information," the entire contents of which are incorporated herein by reference.

[0002] The present application relates to the technical field of identity recognition, and in particular to an identity recognition method, a biometric fingerprint information feature extraction method, an apparatus, a device, and a storage medium. [Background technology]

[0003] With the development of computer technology, the application scenarios of biometric fingerprint recognition technology are expanding, and it is widely applied to scenarios such as attendance records and mobile payments.

[0004] Related technologies typically extract deep features of biometric fingerprint information, reduce the dimension of high-dimensional biometric fingerprint information during the feature extraction process, obtain low-dimensional biometric fingerprint features to describe the biometric fingerprint information, and then recognize the biometric fingerprint information by comparing the biometric fingerprint features. Summary of the Invention [Problem to be solved by the invention]

[0005] The embodiments of the present application provide an identity recognition method, a biometric fingerprint information feature extraction method, an identity recognition device, a biometric fingerprint information feature extraction device, a computer device, and a computer program. [Means for solving the problem]

[0006] According to an embodiment of the present application, there is provided an identity recognition method, the method comprising: acquiring biometric fingerprint information, the biometric fingerprint information describing a biometric fingerprint of a first object; a step of acquiring an overall pattern feature and a local pattern feature by performing a feature extraction process on the biometric pattern information, wherein the overall pattern feature is for describing an overall feature of the biometric pattern information of the first object, and the local pattern feature is for describing a local feature of the biometric pattern information of the first object; performing a fusion process on the global pattern feature and the local pattern feature of the first object to obtain a fused pattern feature of the first object; and performing identity recognition on the first object based on the fused print features of the first object.

[0007] According to an embodiment of the present application, there is provided a method for extracting features of biometric fingerprint information, which is implemented by a biometric fingerprint feature extraction model, the biometric fingerprint feature extraction model including a feature extraction network and a feature fusion network, The method comprises: acquiring biometric fingerprint information, the biometric fingerprint information describing a biometric fingerprint of a first object; calling the feature extraction network and performing feature extraction processing on the biometric pattern information to obtain global pattern features and local pattern features, the global pattern features being for describing global features of the biometric pattern information of the first object, and the local pattern features being for describing local features of the biometric pattern information of the first object; and invoking the feature fusion network to perform a fusion process on the global pattern feature and the local pattern feature of the first object, thereby obtaining a fused pattern feature of the first object.

[0008] According to an embodiment of the present application, there is provided an identity recognition device, comprising: an acquisition module for acquiring biometric print information, the biometric print information describing a biometric print of a first object; an extraction module that acquires global pattern features and local pattern features by performing feature extraction processing on the biometric pattern information, wherein the global pattern features are for describing global features of the biometric pattern information of the first object, and the local pattern features are for describing local features of the biometric pattern information of the first object; a fusion module that performs a fusion process on the global pattern feature and the local pattern feature of the first object to obtain a fused pattern feature of the first object; and a determination module for performing identity recognition on the first object based on the fused print features of the first object.

[0009] According to an embodiment of the present application, the biometric fingerprint feature extraction model is Included A biometric fingerprint information feature extraction device is provided, wherein the biometric fingerprint feature extraction model includes a feature extraction network and a feature fusion network; The device comprises: an acquisition module for acquiring biometric print information, the biometric print information describing a biometric print of a first object; an extraction module that calls the feature extraction network and performs feature extraction processing on the biometric pattern information to obtain global pattern features and local pattern features, wherein the global pattern features are for describing global features of the biometric pattern information of the first object, and the local pattern features are for describing local features of the biometric pattern information of the first object; a fusion module that invokes the feature fusion network to perform a fusion process on the global pattern feature and the local pattern feature of the first object, thereby obtaining a fused pattern feature of the first object.

[0010] According to an embodiment of the present application, there is provided a computing device including a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set, or instruction set, and the at least one instruction, the at least one program, code set, or instruction set, when loaded and executed by the processor, performs the above-described operations. Identity recognition method or A method for extracting features from biometric fingerprint information is realized.

[0011] According to an embodiment of the present application, there is provided a computer-readable storage medium, the computer-readable storage medium storing at least one instruction, at least one program, code set or instruction set, the at least one instruction, the at least one program, code set or instruction set being loaded and executed by a processor to perform the above-described aspects. Identity recognition method or A method for extracting features from biometric fingerprint information is realized.

[0012] According to an embodiment of the present application, there is provided a computer program product or a computer program including computer instructions, the computer instructions being stored in a computer-readable storage medium, and a processor reading and executing the computer instructions from the computer-readable storage medium to perform the method of the above aspect. Identity recognition method or A method for extracting features from biometric fingerprint information is realized. [Brief explanation of the drawings]

[0013] In order to more clearly explain the configuration of the embodiments of the present application, the following briefly introduces the drawings necessary for the description of the embodiments. Obviously, the drawings in the following description only show some embodiments of the present application, and those skilled in the art can derive other drawings from these drawings without creative work.

[0014] [Figure 1] 1 is a schematic diagram of a computer system provided in one exemplary embodiment of the present application. [Figure 2]1 is a flowchart of an identity recognition method provided in one exemplary embodiment of the present application; [Figure 3] 1 is a flowchart of an identity recognition method provided in one exemplary embodiment of the present application; [Figure 4] FIG. 2 is a schematic diagram of an atlas convolution kernel provided in one exemplary embodiment of the present application; [Figure 5] 1 is a flowchart of a method for extracting features from biometric fingerprint information provided in one exemplary embodiment of the present application; [Figure 6] FIG. 2 is a schematic diagram of a ReLU function and a Softplus function provided in one exemplary embodiment of the present application; [Figure 7] 1 is a flowchart of a method for extracting features from biometric fingerprint information provided in one exemplary embodiment of the present application; [Figure 8] 1 is a flowchart of cropping a region of interest of biometric fingerprint information provided in one exemplary embodiment of the present application; [Figure 9] 1 is a flowchart of an identity recognition method provided in one exemplary embodiment of the present application; [Figure 10] 1 is a flowchart of an identity recognition method based on biometric fingerprint information provided in one exemplary embodiment of the present application; [Figure 11] FIG. 2 is a schematic diagram of the construction of a biometric fingerprint feature extraction model provided in one exemplary embodiment of the present application. [Figure 12] 1 is a flowchart of a method for training a biometric fingerprint feature extraction model provided in one exemplary embodiment of the present application; [Figure 13] FIG. 2 is a schematic diagram of feature prediction error determination provided in one exemplary embodiment of the present application; [Figure 14] 1 is a flowchart of a method for training a biometric fingerprint feature extraction model provided in one exemplary embodiment of the present application; [Figure 15] FIG. 1 is a block diagram of a configuration of a biometric fingerprint information feature extraction device provided in one exemplary embodiment of the present application. [Figure 16]FIG. 1 is a block diagram of a configuration of a biometric fingerprint information feature extraction device provided in one exemplary embodiment of the present application. [Figure 17] FIG. 2 is a block diagram of a server configuration provided in one exemplary embodiment of the present application.

[0015] The drawings herein are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application, and together with the specification, serve to explain the principles of the present application. DETAILED DESCRIPTION OF THE INVENTION

[0016] In order to clarify the purpose, configuration, and advantages of the present application, the embodiments of the present application will be described in more detail below with reference to the drawings.

[0017] Illustrative embodiments will now be described in detail, examples of which are illustrated in the drawings. When the following description refers to the drawings, identical numerals in different drawings represent identical or similar elements unless otherwise stated. The embodiments described in the following illustrative examples do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as set forth in the appended claims.

[0018] The terms used in this disclosure are for the purpose of describing particular embodiments only and are not intended to limit the disclosure. As used in this disclosure and the appended claims, the singular forms "a," "the," and "the" are intended to include the plural forms unless the context clearly dictates otherwise. It should also be understood that the term "and / or" as used herein means to include any and all possible combinations of one or more of the associated listed items. It should also be understood that although this disclosure may use terms such as "first," "second," etc. to describe various information, such information should not be limited to these terms. These terms are used only to distinguish between the same type of information. For example, a first parameter could be referred to as a second parameter, and similarly, a second parameter could be referred to as a first parameter, without departing from the scope of this disclosure. Depending on the context, the word "when" as used herein can be interpreted as "when," "when," or "in response to a determination."

[0019] It should be noted that all user information (including, but not limited to, user device information and user personal information) and data (including, but not limited to, analytical data, stored data, and display data) related to this application are information and data authorized by the user or fully authorized by each party, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, all biometric pattern information related to this application was obtained with sufficient authorization.

[0020] 1 is a schematic diagram of a computer system provided in one exemplary embodiment of the present application. The computer system may be implemented as a system architecture for a biometric fingerprint information feature extraction method. The computer system may include a terminal 100 and a server 200.

[0021] The terminal 100 may be, for example, an electronic device such as a mobile phone, a tablet computer, an in-vehicle terminal (in-vehicle device), a wearable device, a personal computer (PC), an access control device, or an unmanned vending machine. A client that executes a target application may be installed in the terminal 100. The target application may be a mobile payment, attendance, or identity authentication application that supports biometric fingerprint information authentication, or may be another application that provides a biometric fingerprint information feature extraction function, but this application is not limited thereto. Note that, in this application, the form of the target application is not limited and includes, but is not limited to, an application (App) or applet installed on the terminal 100, and may be in the form of a web page.

[0022] The server 200 may be an independent physical server, a server cluster or a distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services. The server 200 may be a backend server of the target application described above, providing backend services to clients of the target application.

[0023] In the biometric fingerprint information feature extraction method provided in the embodiments of the present application, each step may be performed by a computer device. The computer device refers to an electronic device capable of calculating, processing, and storing data. Taking the method implementation environment shown in FIG. 1 as an example, the terminal 100 may execute the biometric fingerprint information feature extraction method (e.g., a client running a target application installed on the terminal 100 executes the biometric fingerprint information feature extraction method), the server 200 may execute the biometric fingerprint information feature extraction method, or the terminal 100 and the server 200 may execute the biometric fingerprint information feature extraction method through mutual cooperation, but this is not limited to this. Furthermore, the configurations of the present application may be combined with blockchain technology. For example, some of the data related to the biometric fingerprint information feature extraction method disclosed in the present application (e.g., biometric fingerprint information, identity information corresponding to the biometric fingerprint information, etc.) may be stored on a blockchain. Communication between the terminal device 100 and the server 200 is possible via a network. The network is For example, a wired or wireless network.

[0024] In related art, recognition ability is significantly reduced for highly similar biometric fingerprint information, such as the biometric fingerprint information of twins. Based on this, the present application provides a method, device, equipment, and storage medium for extracting biometric fingerprint information features. Next, the biometric fingerprint information feature extraction method is introduced in the following examples.

[0025] 2 shows a flowchart of an identity recognition method provided in one exemplary embodiment of the present application, which may be performed by a computer device. The method includes the following steps:

[0026] In step 510, biometric fingerprint information is obtained.

[0027] The biometric information is for describing the biometric print of the first object. In the present embodiment, the biometric print type includes, but is not limited to, at least one of a finger print, a palm print, an eye print, a face print, a vein print, a lip print, and an oral print. The present application does not impose any restrictions on the biometric print type.

[0028] In the embodiments of the present application, the biometric fingerprint collection method includes, but is not limited to, at least one of an optical collection method (e.g., biometric fingerprint photography), a capacitive collection method (e.g., using a capacitive biometric fingerprint collection module), and a biometric radio frequency collection method (e.g., using ultrasound to collect biometric fingerprints). The present application does not impose any restrictions on the biometric fingerprint collection method.

[0029] In step 520, a feature extraction process is performed on the biometric pattern information to obtain global and local pattern features.

[0030] The global pattern feature is for describing the global feature of the biometric pattern information of the first object, and the local pattern feature is for describing the local feature of the biometric pattern information of the first object. In the embodiment of the present application, the local pattern feature may be acquired based on the global pattern feature, or the local pattern feature and the global pattern feature may be acquired independently of each other, but no limitation is imposed thereon.

[0031] It should be noted that the biometric feature extraction process may involve invoking a biometric feature extraction model, manually annotating the biometric feature, or computing the biometric feature, and the present application does not impose any restrictions thereon.

[0032] As an example, a case will be taken where feature extraction processing is performed on biometric pattern information using a biometric pattern feature extraction model. The biometric pattern feature extraction model includes a feature extraction network. By calling the feature extraction network and performing feature extraction processing on the biometric pattern information, global pattern features and local pattern features are obtained. The global pattern feature is used to describe the global features of the biometric pattern information of a first object, and the local pattern feature is used to describe the local features of the biometric pattern information of the first object.

[0033] In step 530, a fusion process is performed on the global pattern feature and the local pattern feature of the first object to obtain a fused pattern feature of the first object.

[0034] For example, the fused print feature is for describing the global feature and the local feature of the biometric print information of the first object. In the embodiment of the present application, the fusion process for the global feature and the local feature may be performed in a manner including, but not limited to, at least one of orthogonal fusion, convolutional fusion, and self-adaptive feature fusion.

[0035] It should be noted that, similar to step 520 above, the present application does not impose any restrictions on the manner in which the fusion process is performed. Take the case in which the fusion process is performed on the global print feature and the local print feature using a biometric print feature extraction model as an example. The biometric print feature extraction model includes a feature fusion network. The feature fusion network is invoked to perform the fusion process on the global print feature and the local print feature of the first object, thereby obtaining a fused print feature of the first object.

[0036] In step 540, identity recognition is performed on the first object based on the fused print features of the first object.

[0037] As described above, in the method provided in this embodiment, local pattern features are acquired by performing feature extraction processing on biometric pattern information, and attention is paid to local detailed features of the biometric pattern. By combining the local pattern features and the overall pattern features, it is possible to obtain a high-quality image based on the pattern features. bioprint This will improve the ability to describe information. bioprint The dimension that locally describes the information is expanded, ensuring recognition accuracy when faced with highly similar biometric information.

[0038] Next, the feature extraction process will be introduced. Figure 3 shows a flowchart of an identity recognition method provided in one exemplary embodiment of the present application. The method may be performed by a computer device. That is, in some embodiments, step 520 in the embodiment shown in Figure 2 may be realized as the following two steps:

[0039] In step 522, hidden layer feature extraction is performed on the biometric print information to obtain texture hidden layer features of the first object.

[0040] The texture hidden layer features are bioprint For describing hidden layer representation of information, illustratively, the representation method of texture hidden layer features includes, but is not limited to, at least one of a feature vector, a feature matrix, and a feature value.

[0041] It should be noted that, similar to the above step 520, the present application does not impose any restrictions on the implementation method of hidden layer feature extraction. Take the case where hidden layer feature extraction is performed on biometric print information by a biometric print feature extraction model as an example. The biometric print feature extraction model includes a feature extraction network, which includes a first residual convolution block. The first residual convolution block is called to perform hidden layer feature extraction on the biometric print information to obtain texture hidden layer features of the first object. The texture hidden layer features are: bioprint It is intended to describe hidden layer representations of information.

[0042] For example, the hidden layer feature extraction method for biometric fingerprint information is a feature extraction method based on feature dimension reduction. In the process of performing hidden layer feature extraction using the biometric fingerprint feature extraction model, the hidden layer feature extraction is realized by calling the first residual convolution block.

[0043] In some embodiments, the first residual convolution block includes at least one of, but is not limited to, Convolutional Neural Networks (CNNs), Efficient Networks, Transformer Networks, Deep Residual Networks (ResNets), and Inception-Residual Networks (Inception-Resnets).

[0044] In step 524, a first feature extraction method is used to perform feature extraction on the texture hidden layer features to obtain the overall pattern features of the first object.

[0045] The first feature extraction method is a feature extraction method based on feature dimension reduction. For example, the feature dimension is reduced by a residual convolution method.

[0046] As with step 520 above, the present application does not impose any restrictions on the method of performing feature extraction. Take the example of performing hidden layer feature extraction on biometric information using a biometric feature extraction model. The biometric feature extraction model includes a feature extraction network, which includes a second residual convolution block. The second residual convolution block is called to perform feature extraction on the texture hidden layer features using a first feature extraction method to obtain the overall biometric feature of the first object. The first feature extraction method is a feature extraction method based on feature dimension reduction.

[0047] In step 526, the local pattern feature of the first object is obtained by performing feature extraction on the texture hidden layer feature using a second feature extraction method.

[0048] The second feature extraction method is a feature extraction method based on expanding the feature receptive field. For example, the feature receptive field is expanded using an atlas convolution method.

[0049] Illustratively, the feature receptive field is for indicating the information dimension of the texture hidden layer feature indicated by each feature value of the local pattern feature. Illustratively, in the process of performing feature extraction on the first original image using the first feature image, the feature receptive field represents the dimension of the original image pixel indicated by one image feature pixel in the first feature image.

[0050] Similar to step 520 above, the present application does not impose any restrictions on the implementation method of feature extraction. Take the case where hidden layer feature extraction is performed on biometric information using a biometric feature extraction model as an example. The biometric feature extraction model includes a feature extraction network, which includes an atlas convolution block. The atlas convolution block is called to perform feature extraction on the texture hidden layer features using a second feature extraction method to obtain local biometric features of the first object. The second feature extraction method is a feature extraction method based on feature receptive field expansion.

[0051] In some embodiments of the present application, step 526 may be implemented as two sub-steps as follows.

[0052] In sub-step 1, the texture hidden layer features are processed to obtain n local sub-features of the first object using n second feature extraction methods, respectively.

[0053] n is a positive integer, and the n second feature extraction methods have different feature receptive fields.

[0054] As with step 520 above, the present application does not impose any restrictions on the implementation method of feature extraction. Take for example a case where hidden layer feature extraction is performed on biometric information using a biometric print feature extraction model. The biometric print feature extraction model includes a feature extraction network, which includes an atlas convolution block, which includes n atlas convolution layers. The n atlas convolution layers are called to process the texture hidden layer features, and n second feature extraction methods respectively obtain n local sub-features of the first object. 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 atlas convolution layer.

[0055] In sub-step 2, the n local sub-features are combined to obtain a local print feature of the first object.

[0056] When n is greater than 1, the local sub-features may be directly combined from end to end, or may be overlapped and mathematically combined, although this embodiment does not impose any restrictions on this.

[0057] If n is equal to 1, determine the local sub-feature as the local feature of the first object.

[0058] Illustratively, Figure 4 shows a schematic diagram of an atlas convolution kernel provided in one exemplary embodiment of the present application. kernel The schematic diagram shows an atlas convolution kernel with a convolution kernel scale of 3, an atlas convolution kernel with a convolution kernel scale of 6, and an atlas convolution kernel with a convolution kernel scale of 9. Here, the feature receptive field of the atlas convolution kernel with a convolution kernel scale of 3 has dimensions of 5x5, the feature receptive field of the atlas convolution kernel with a convolution kernel scale of 6 has dimensions of 11x11, and the feature receptive field of the atlas convolution kernel with a convolution kernel scale of 9 has dimensions of 17x17.

[0059] As described above, in the method provided in this embodiment, by performing feature extraction processing on biometric pattern information, local pattern features are acquired by a feature extraction method based on the expansion of the feature receptive field, and attention is paid to the local detailed features of the biometric pattern. By combining the local pattern features and the overall pattern features, it is possible to obtain a high-quality image based on the pattern features. bioprint This will improve the ability to describe information. bioprint The dimension that locally describes the information is expanded, ensuring recognition accuracy when faced with highly similar biometric information.

[0060] Next, the self-attention mechanism process will be introduced. Figure 5 shows a flowchart of a biometric fingerprint feature extraction method provided in one exemplary embodiment of the present application. The method may be performed by a computer device. That is, in some embodiments, in addition to the embodiment shown in Figure 2, The method comprises: Further included is step 528.

[0061] In step 528, the local pattern features are subjected to self-attention processing to obtain updated local pattern features.

[0062] For example, a self-attention mechanism process is performed on the local pattern features to strengthen the weights of the local effective feature regions, thereby obtaining updated local pattern features.

[0063] Illustratively, in some embodiments of the present application, step 528 may be implemented as three sub-steps as follows: the self-attention mechanism process includes an activation process and a regularization process.

[0064] In sub-step 1, second activation features are obtained by performing activation processing on the local pattern features.

[0065] Illustratively, the activation function for performing activation processing on the local print features includes, but is not limited to, at least one of a Sigmoid function, a Hyperbolic Tangent (Tanh) function, a Rectified Linear Unit (ReLU) function, a Softmax function, and a Softplus function. Figure 6 shows a schematic diagram of the ReLU function and the Softplus function provided in one exemplary embodiment of the present application.

[0066] For example, by performing activation processing on the local pattern features using the ReLU function and the Softplus function, the second activation features are obtained as shown in the following formula, for example.

number

[0067] For example, the ReLU function can increase the nonlinear relationships between each neural network layer in the biometric fingerprint feature extraction model and prevent simple linear relationships from appearing. The ReLU function outputs some local fingerprint features as 0 through maximum value calculation, which makes the network sparsity and reduces the interdependence between local fingerprint features, thereby mitigating the overfitting problem. The Softplus function has the same function as the ReLU function and is a smoothed version of the ReLU function.

[0068] In sub-step 2, a regularization process is performed on the local pattern features to obtain second regularized features.

[0069] For example, the method of performing regularization processing on the local pattern feature may be L1 parameter regularization or L2 parameter regularization. For example, by performing regularization processing on the local pattern feature using L2 parameter regularization, a second regularized feature is obtained, for example, as shown in the following formula:

number

[0070] For example, L2 parameter regularization controls the value of the second regularization feature between 0 and 1 by calculating the Euclidean distance, effectively preventing overfitting of the second regularization feature and improving the generalization ability of the model.

[0071] In sub-step 3, the second activation feature and the second regularization feature are multiplied to obtain the updated local pattern feature of the first object.

[0072] In some embodiments of the present application, before sub-step 1, The method comprises: The method further includes performing a standardization process on the local print features to obtain second standardized features.

[0073] Accordingly, sub-step 1 performs the following for the second standardized feature: activation By processing, the second activation Sub-step 2 is realized by performing a regularization process on the second standardized features to obtain second regularized features.

[0074] The standardization process is performed on the local pattern features to ensure that the second standardized features are distributed within a certain interval, which provides a basis for the model to quickly achieve convergence results. For example, the standardization process on the local pattern features is as follows:

number

[0075] In step 530a, a fusion process is performed on the global pattern feature and the updated local pattern feature of the first object, thereby obtaining a fused pattern feature of the first object.

[0076] For example, the fused print feature is for describing the global feature and the local feature of the biometric print information of the first object. In the embodiment of the present application, the fusion process for the global feature and the local feature may be performed in a manner including, but not limited to, at least one of orthogonal fusion, convolutional fusion, and self-adaptive feature fusion.

[0077] For example, the fused pattern feature of the first object is obtained by performing a fusion process on the global pattern feature and the updated local pattern feature of the first object using an orthogonal fusion method.

number

[0078] As described above, in the method provided in this embodiment, local pattern features are obtained by performing feature extraction processing on biometric pattern information, and attention to local detailed features of the biometric pattern is increased. The weight of the local effective feature area is strengthened by self-attention processing, and the local pattern features and the global pattern features are combined to obtain a result based on the pattern features. bioprint This will improve the ability to describe information. bioprint The dimension that locally describes the information is expanded, ensuring recognition accuracy when faced with highly similar biometric information.

[0079] Next, the pre-processing will be introduced. Figure 7 shows a flowchart of a biometric fingerprint feature extraction method provided in one exemplary embodiment of the present application. The method may be performed by a computer device. That is, in some embodiments, in addition to the embodiment shown in Figure 2, The method comprises: Further included is step 512.

[0080] In step 512, preprocessing is performed on the biometric pattern information to obtain updated biometric pattern information.

[0081] Exemplarily, the pre-processing method includes at least one of performing an interpolation process on the biometric pattern information (exemplarily, performing an interpolation process on the biometric pattern information is used to change the dimension of the biometric pattern information) and cropping a region of interest (ROI) of the biometric pattern information (exemplarily, cropping a region of interest of the biometric pattern information is used to identify a key region of interest), but is not limited thereto.

[0082] Illustratively, performing an interpolation process on the biometric print information is used to change the dimension of the biometric print information and convert the biometric print information into a target dimension, for example, performing an interpolation process on the biometric print information to convert the biometric print information into a feature matrix with dimensions 224×224.

[0083] For example, this embodiment does not impose any restrictions on the method for extracting the region of interest of the biometric fingerprint information, and the region of interest may be determined by a calculation rule or a region of interest extraction network.

[0084] 8 is a schematic diagram of cropping a region of interest from biometric fingerprint information provided in one exemplary embodiment of the present application. Taking the biometric fingerprint information as palmprint 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 palmprint image information. Here, the first key point A is the gap between the second and third fingers on the palm, the second key point B is the gap between the third and fourth fingers on the palm, and the third key point C is the gap between the third and fourth fingers on the palm. key Point C is the gap between the fourth and fifth fingers on the palm.

[0085] A rectangular coordinate system is created for the palm print image information. Here, the direction from the third key point C to the first key point A is determined as the positive direction of the x-axis, and the intersection point E of the perpendicular line passing through the second key point B with the x-axis direction is determined. Perpendicular The direction from the intersection point E to the second key point B is determined as the positive direction of the y-axis, and the perpendicular line is extended from the intersection point E along the negative direction of the y-axis to determine the center point D of the palm, and the region of interest of the palmprint image information is determined with the center point D of the palm as the center. Here, the region of interest of the palmprint image information is usually square, but other shapes such as rectangle, circle, fan shape, etc. are not excluded. For example, the center point D of the palm and Perpendicular The distance to intersection point E is 6 / 5 times the distance from the first key point A to the third key point C, and the region of interest in the palm print image information is a square region centered on the center point D of the palm, with side lengths 2 / 3 times the distance from the first key point A to the third key point C.

[0086] In step 520a, feature extraction processing is performed on the updated biometric pattern information to obtain overall pattern features and local pattern features.

[0087] The global pattern feature is for describing the global feature of the biometric pattern information of the first object, and the local pattern feature is for describing the local feature of the biometric pattern information of the first object. In the embodiment of the present application, the local pattern feature may be acquired based on the global pattern feature, or the local pattern feature and the global pattern feature may be acquired independently of each other, but no limitation is imposed thereon.

[0088] As described above, in the method provided in this embodiment, local pattern features are obtained by performing feature extraction processing on biometric pattern information, and attention is paid to local detailed features of the biometric pattern. The basis for obtaining the local pattern features is determined by pre-processing, and the local pattern features and the overall pattern features are combined to obtain the biometric pattern features. bioprint This will improve the ability to describe information. bioprint The dimension that locally describes the information is expanded, ensuring recognition accuracy when faced with highly similar biometric information.

[0089] Next, identity determination will be introduced. Figure 9 shows a flowchart of an identity recognition method provided in one exemplary embodiment of the present application. The method may be performed by a computer device. That is, in some embodiments, step 540 in the embodiment shown in Figure 2 may be specifically step 5401.

[0090] In step 5401, the identity determination result of the first object is obtained by comparing the fused print feature of the first object with the sample print feature.

[0091] The sample fingerprint feature is for describing the biometric fingerprint information feature of the target object, and the identity determination result of the first object is for indicating whether the first object and the target object belong to the same identity.

[0092] Illustratively, the method for comparing the fused print feature of the first object with the sample print feature includes, but is not limited to, at least one of a cosine similarity calculation, a Euclidean distance calculation, a covariance distance calculation, and a Chebyshev distance calculation.

[0093] In some embodiments of the present application, step 540 may be implemented as the sub-steps of: calculating the cosine similarity between the fused print feature and the sample print feature; obtaining an identity judgment result of the first object being the target object if the cosine similarity satisfies the identity judgment condition; and obtaining an identity judgment result of the first object being not the target object if the cosine similarity does not satisfy the identity judgment condition.

[0094] Illustratively, this involves calculating the cosine similarity as follows:

number

[0095] Exemplarily, the identity criterion is that the cosine similarity is greater than a target threshold, where the target threshold may be set empirically or obtained through a training process.

[0096] As described above, in the method provided in this embodiment, local biometric features are obtained by performing feature extraction processing on biometric information, and attention is paid to local detailed features of the biometric feature, and identity is determined by comparing the biometric features. bioprintThe dimension that locally describes the information is expanded, ensuring recognition accuracy when faced with highly similar biometric information.

[0097] Next, the biometric fingerprint feature extraction method will be described as being applied to the field of mobile payment. Figure 10 shows a flowchart of an identity recognition method based on biometric fingerprint information provided in one exemplary embodiment of the present application. The method may be performed by a computer device. The method includes the following steps:

[0098] In step 610, the terminal payment device obtains biometric pattern information of the first object.

[0099] The biometric information describes the biometric print of the first object. In the present embodiment, the type of biometric print includes, but is not limited to, at least one of a fingerprint, a palmprint, an eyeprint, a faceprint, a vein print, a lipprint, and an oral cavity print. In the present embodiment, the biometric print is a palmprint, i.e., the biometric print information is image information of the palmprint.

[0100] In step 620, the terminal payment system detects the biometric information to obtain finger gap key points.

[0101] Illustratively, obtaining the finger gap key points is used to create a Cartesian coordinate system in the palmprint image information, which then crops out a region of interest in the palmprint image information.

[0102] 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, but this embodiment is not limited thereto.

[0103] Similarly, no restrictions are placed on the execution method of steps 630 to 670.

[0104] In step 630, the terminal payment system crops the region of interest of the biometric print information based on the finger gap key points.

[0105] The terminal payment system creates a Cartesian coordinate system based on the finger gap key points and cuts out the region of interest of the biometric fingerprint information.

[0106] In step 642, the terminal payment system performs global feature extraction on the region of interest of the biometric fingerprint information to obtain the global fingerprint feature of the first object.

[0107] For example, global feature extraction for a region of interest of biometric fingerprint information is performed by adopting a feature extraction method based on feature dimension reduction, specifically, by invoking a residual convolution block to realize global feature extraction and obtain the global fingerprint feature of the first object.

[0108] In step 644, the terminal payment system obtains the local fingerprint features of the first object by performing local feature extraction on the region of interest of the biometric fingerprint information using three different receptive field feature extraction methods.

[0109] For example, local feature extraction is performed on the region of interest of the biometric fingerprint information by adopting a feature extraction method based on feature receptive field expansion, specifically by invoking an atlas convolution block to realize local feature extraction and obtain the local fingerprint features of the first object.

[0110] In step 646, the terminal payment system performs self-attention processing on the local print features to obtain updated local print features.

[0111] The self-attention mechanism processing is performed on the local pattern features to strengthen the weights of the local effective feature regions, thereby obtaining updated local pattern features.

[0112] Illustratively, in some embodiments of the present application, the self-attention mechanism process includes a standardization process, an activation process, and a regularization process.

[0113] In step 650, the terminal payment system performs a fusion process on the global print feature and the updated local print feature to obtain a fused print feature of the first object.

[0114] For example, the fused print feature is for describing the global feature and the local feature of the biometric print information of the first object. Specifically, the fused print feature of the first object is obtained by performing a fusion process on the global print feature and the updated local print feature using an orthogonal fusion method.

[0115] In step 660, the terminal payment system calculates the cosine similarity between the fused print feature and the sample print feature.

[0116] Illustratively, the cosine similarity indicates the similarity between the fused print feature and the sample print feature by the cosine angle between the vectors.

[0117] In step 670, the terminal payment system determines the sample object corresponding to the sample with the highest cosine similarity as the identity recognition result of the first object.

[0118] The cosine similarities are sorted in descending order. The sample pattern feature with the highest cosine similarity indicates the highest similarity between the sample pattern feature and the fused pattern feature. The sample object corresponding to the sample pattern feature is determined as the identity recognition result of the first object.

[0119] As described above, in the method provided in this embodiment, local pattern features are acquired by performing feature extraction processing on biometric pattern information, and attention is paid to local detailed features of the biometric pattern. By combining the local pattern features and the overall pattern features, it is possible to obtain a high-quality image based on the pattern features. bioprint This will improve the ability to describe information. bioprintThe dimension that locally describes the information is expanded, ensuring recognition accuracy when faced with highly similar biometric information.

[0120] Next, a biometric fingerprint feature extraction model will be described. In the embodiment shown in Fig. 2, a method for performing identity recognition by the biometric fingerprint feature extraction model will be described. Fig. 11 shows a schematic diagram of the configuration of the biometric fingerprint feature extraction model provided in one exemplary embodiment of the present application.

[0121] The biometric print feature extraction model includes a feature extraction network 330, a self-attention network 350, and a feature fusion network 360.

[0122] Region of interest pattern information 310a is obtained by performing a region of interest cropping process 322 on the biometric pattern feature 310. Interpolation process 324 is performed on the region of interest pattern information 310a to obtain interpolated pattern information 310b. Illustratively, the interpolated pattern information 310b is a feature matrix of dimensions 224×224.

[0123] The feature extraction network 330 includes a first residual convolution block 332, a second residual convolution block 334, and an atlas convolution block 336.

[0124] The first residual convolution block 332 is called to process the interpolated pattern information 310b to obtain texture hidden layer features. For example, the texture hidden layer features are feature matrices with dimensions of 28×28. Here, the first residual convolution block 332 includes three residual convolution layers, and the dimensions of the feature matrices output from the residual convolution layers are one-fourth the dimensions of the input feature matrices. That is, when the first residual convolution layer is called to process the interpolated pattern information 310b, the dimensions of the output feature matrix are 112×112; when the second residual convolution layer is called to process the same feature matrix, the dimensions of the output feature matrix are 56×56; and when the third residual convolution layer is called to process the same feature matrix, the dimensions of the output feature matrix are 28×28.

[0125] The second residual convolution block 332 is called to process the texture hidden layer features to obtain the global pattern feature 342. Illustratively, the global pattern feature 342 is a feature matrix with dimensions 14x14.

[0126] The atlas convolution block 336 includes a first atlas convolution layer 336a, a second atlas convolution layer 336b, and a third atlas convolution layer 336c.

[0127] The first atlas convolutional layer 336a is called to process the texture hidden layer features to obtain first local sub-features, and the atlas convolutional kernel scale of the first atlas convolutional layer 336a is 3. The second atlas convolutional layer 336b is called to process the texture hidden layer features to obtain second local sub-features, and the atlas convolutional kernel scale of the second atlas convolutional layer 336b is 6. The third atlas convolutional layer 336c is called to process the texture hidden layer features to obtain third local sub-features, and the atlas convolutional kernel scale of the third atlas convolutional layer 336c is 9.

[0128] The first local sub-feature, the second local sub-feature, and the third local sub-feature are combined to obtain an initial local print feature 344 .

[0129] The self-attention network 350 includes a normalization layer 352, an activation layer 354, and a regularization layer 356.

[0130] By calling the normalization layer to process the initial local pattern features 344, it is ensured that the updated local pattern features 344a are distributed within a certain interval, which provides a basis for the model to quickly achieve convergence results.

[0131] The activation layer 354 and the regularization layer 356 are called to perform activation and regularization processes on the standardized local pattern features, respectively, and the results are multiplied to obtain updated local pattern features 344a.

[0132] The updated local pattern feature 344a is multiplied by the initial local pattern feature 344 to obtain a local pattern feature 344c.

[0133] The feature fusion network 360 is invoked to perform fusion processing on the global pattern feature 342 and the local pattern feature 344c of the first object, thereby obtaining a fused pattern feature 346 of the first object.

[0134] The cosine similarity 352 between the fused print feature 346 and the sample print feature 348 is calculated.

[0135] Based on cosine similarity 352, the first Object The identity determination result 370 is determined.

[0136] That is, if the cosine similarity 352 satisfies the identity determination condition, an identity determination result 370 of the first object is obtained, which is that the first object is the target object.

[0137] If the cosine similarity 352 does not satisfy the identity determination condition, an identity determination result 370 of the first object is obtained, which is that the first object is not the target object.

[0138] In an embodiment of the present application, there is provided a feature extraction method for biometric print information. The method is executed by a biometric print feature extraction model of a computing device, the biometric print feature extraction model including a feature extraction network and a feature fusion network. The method includes the steps of acquiring biometric print information, the biometric print information describing a biometric print of a first object; invoking the feature extraction network to perform a feature extraction process on the biometric print information to acquire a global print feature and local print features, the global print feature describing a global feature of the biometric print information of the first object and the local print features describing a local feature of the biometric print information of the first object; and invoking the feature fusion network to perform a fusion process on the global print feature and the local print features of the first object to acquire a fused print feature of the first object.

[0139] Next, we introduce the training of the biometric print feature extraction model.

[0140] 12 shows a flowchart of a method for training a biometric fingerprint feature extraction model provided in one exemplary embodiment of the present application. The method may be performed by a computer device. The method includes the following steps:

[0141] In step 710, sample print information and sample print features are obtained.

[0142] The sample print information is for describing the biometric print of the sample object, and the sample print features are biometric print features corresponding to the sample print information of the sample object.

[0143] In the embodiments of the present application, the type of biometric fingerprint includes, but is not limited to, at least one of a finger print, a palm print, an eye print, a face print, a blood vessel print, a lip print, and an oral cavity print. The present application does not impose any restrictions on the type of biometric fingerprint.

[0144] In step 720, the initial biometric print feature extraction model is invoked to process the sample print information and output predicted fused print features of the sample object.

[0145] Illustratively, the predicted fusion print features are for describing global and local features of biometric print information of a sample object.

[0146] In step 730, the predicted fused print feature is compared with the sample print feature to obtain the feature prediction error.

[0147] The sample print features are biometric print features corresponding to the sample print information of the sample object.

[0148] Illustratively, the feature prediction error is a loss function of a softmax function between the predicted fused print feature and the sample print feature.

[0149] Specifically, this involves obtaining the feature prediction error as follows:

number

[0150] In step 740, the feature prediction error is used to perform back-propagation training on the initial biometric feature extraction model to obtain a biometric feature extraction model.

[0151] The purpose of performing backpropagation training on the initial biometric fingerprint feature extraction model is to minimize the feature prediction error between the predicted fusion biometric fingerprint feature and the sample biometric fingerprint feature. By training the initial biometric fingerprint feature extraction model, a biometric fingerprint feature extraction model is obtained.

[0152] As described above, in the method provided in this embodiment, the local and global pattern features are integrated by performing backpropagation training on the initial biometric pattern feature extraction model, and the biometric pattern feature is extracted. bioprint This lays the foundation for improving the ability to describe information. Taking palmprint image data as an example, the biometric print feature extraction model achieves good training results. Table 1 shows the results of identity recognition on a set of clear and blurred palmprint images, including 40 pairs of twins. For 3,600 sample pairs, each pair representing the left and right hands of a pair of twins, the biometric print feature extraction model made no errors when faced with clear palmprint images, but only one error when faced with blurred palmprint images. This represents a significant improvement over the 37 and 46 errors reported by conventional models in the related art.

[0153] [Table 1]

[0154] 14 shows a flowchart of a method for training a biometric fingerprint feature extraction model provided in one exemplary embodiment of the present application. The method may be performed by a computer device. The method includes the following steps:

[0155] In step 750, the sample pattern information and the classification label of the sample pattern information are obtained.

[0156] The sample print information is for describing the biometric print of the sample object, and the classification label is for indicating the sample object.

[0157] In the embodiments of the present application, the type of biometric fingerprint includes, but is not limited to, at least one of a finger print, a palm print, an eye print, a face print, a blood vessel print, a lip print, and an oral cavity print. The present application does not impose any restrictions on the type of biometric fingerprint.

[0158] In step 760, the initial biometric print feature extraction model is invoked to process the sample print information and output predicted fused print features of the sample object.

[0159] Illustratively, the predicted fusion print features are for describing global and local features of biometric print information of a sample object.

[0160] In step 770, a label classification process is performed on the predicted fused print features to obtain a predicted label for the sample object.

[0161] For example, in this embodiment, the label classification process may be realized by calling a label prediction model to realize the label classification process of the predicted fusion print feature, or may be performed by a manual annotation method, or may obtain the predicted label of the sample object by calculating the predicted fusion print feature. This embodiment does not impose any restrictions on the implementation method of the label classification process.

[0162] In step 780, the predicted label is compared with the classification label to obtain the label prediction error.

[0163] The classification label is for indicating the sample object. Illustratively, the label prediction error includes at least one of a zero-one loss function, an absolute value loss function, and a logarithmic loss function between the predicted label and the classification label.

[0164] In step 790, the label prediction error is used to perform backpropagation training on the initial biometric feature extraction model to obtain a biometric feature extraction model.

[0165] The purpose of performing backpropagation training on the initial biometric fingerprint feature extraction model is to minimize the feature prediction error between the predicted label and the classification label. By training the initial biometric fingerprint feature extraction model, a biometric fingerprint feature extraction model is obtained.

[0166] As described above, the method provided in this embodiment performs backpropagation training on the initial biometric fingerprint feature extraction model to integrate the local and global fingerprint features, thereby obtaining a biometric fingerprint feature. bioprint Establish a foundation for improving your ability to write about information.

[0167] As will be understood by those skilled in the art, the above embodiments may be implemented independently, or the above embodiments may be freely combined to form new embodiments, thereby realizing the biometric pattern information feature extraction method of the present application.

[0168] 15 shows a block diagram of an identity recognition device provided in an exemplary embodiment of the present application, the device including: an acquisition module 810 for acquiring biometric print information, where the biometric print information describes a biometric print of a first object; an extraction module 820 for performing a feature extraction process on the biometric print information to acquire global print features and local print features, where the global print features describe global features of the biometric print information of the first object and the local print features describe local features of the biometric print information of the first object; a fusion module 830 for performing a fusion process on the global print features and the local print features of the first object to acquire a fused print feature of the first object; and a determination module 850 for performing identity recognition on the first object based on the fused print feature of the first object.

[0169] In some embodiments of the present application, the extraction module 820 further performs hidden layer feature extraction on the biometric print information to obtain texture hidden layer features of the first object; performs feature extraction on the texture hidden layer features using a first feature extraction method to obtain the global print features of the first object; and performs feature extraction on the texture hidden layer features using a second feature extraction method to obtain the local print features of the first object, and the texture hidden layer features are bioprint The first feature extraction method is for describing a hidden layer representation of information, and the second feature extraction method is a feature extraction method based on feature dimension reduction, and the second feature extraction method is a feature extraction method based on feature receptive field expansion.

[0170] In some embodiments of the present application, the extraction module 820 further processes the texture hidden layer features, obtains n local sub-features of the first object using n (n is a positive integer) second feature extraction methods, respectively, and obtains the local pattern feature of the first object by combining the n local sub-features, where the n second feature extraction methods have different feature receptive fields.

[0171] In some embodiments of the present application, the device further includes a processing module 840 that performs self-attention processing on the local pattern features to obtain updated local pattern features, and the fusion module 830 further performs fusion processing on the global pattern features and the updated local pattern features of the first object to obtain fused pattern features of the first object.

[0172] In some embodiments of the present application, the self-attention mechanism process includes an activation process and a regularization process.

[0173] The processing module 840 further performs the activation process on the local pattern feature to obtain a second activation feature, performs the regularization process on the local pattern feature to obtain a second regularization feature, and multiplies the second activation feature by the second regularization feature to obtain the updated local pattern feature of the first object.

[0174] In some embodiments of the present application, the device further includes a processing module 840 for performing pre-processing on the biometric fingerprint information to obtain updated biometric fingerprint information, where the pre-processing manner includes, but is not limited to, at least one of: performing interpolation on the biometric fingerprint information; and cropping a region of interest of the biometric fingerprint information.

[0175] The extraction module 820 further performs feature extraction processing on the updated biometric pattern information to obtain global and local pattern features.

[0176] In some embodiments of the present application, the device further includes a determination module 850 for obtaining an identity determination result of the first object by comparing the fused print feature of the first object with a sample print feature, where the sample print feature is for describing the biometric print information feature of a target object.

[0177] In some embodiments of the present application, the judgment module 850 further calculates the cosine similarity between the fused print feature and the sample print feature, and if the cosine similarity satisfies the identity judgment condition, obtains the identity judgment result of the first object that it is the target object; if the cosine similarity does not satisfy the identity judgment condition, obtains the identity judgment result of the first object that it is not the target object.

[0178] 16 shows a block diagram of a biometric information feature extraction device provided in one exemplary embodiment of the present application. The device is configured to extract biometric information features based on a biometric feature extraction model. Includes The biometric print feature extraction model includes a feature extraction network and a feature fusion network, and the apparatus includes: an acquisition module 860 that acquires biometric print information, the biometric print information describing the biometric print of a first object; an extraction module 870 that invokes the feature extraction network to perform feature extraction processing on the biometric print information to acquire global print features and local print features, the global print feature describing a global feature of the biometric print information of the first object, and the local print features describing a local feature of the biometric print information of the first object; and a fusion module 880 that invokes the feature fusion network to perform a fusion processing on the global print feature and the local print features of the first object to acquire a fused print feature of the first object.

[0179] In some embodiments of the present application, the feature extraction network includes a first residual convolution block, a second residual convolution block, and an atlas convolution block, and the extraction module 870 further calls the first residual convolution block to perform hidden layer feature extraction on the biometric print information to obtain texture hidden layer features of the first object; calls the second residual convolution block to perform feature extraction on the texture hidden layer features in a first feature extraction manner to obtain the global print features of the first object; and calls the atlas convolution block to perform feature extraction on the texture hidden layer features in a second feature extraction manner to obtain the local print features of the first object, and the texture hidden layer features are bioprint The first feature extraction method is for describing a hidden layer representation of information, and the second feature extraction method is a feature extraction method based on feature dimension reduction, and the second feature extraction method is a feature extraction method based on feature receptive field expansion.

[0180] In some embodiments of the present application, the atlas convolution block includes n atlas convolution layers (n is a positive integer), and the extraction module 870 further invokes the n atlas convolution layers to process the texture hidden layer features, obtains n local sub-features of the first object using n second feature extraction methods, respectively, and combines the n local sub-features to obtain the local pattern feature of the first object, where the n second feature extraction methods have different feature receptive fields, and the feature receptive fields are determined based on the convolution kernel scale of the atlas convolution layer.

[0181] In some embodiments of the present application, the biometric print feature extraction model is trained by: acquiring sample print information and sample print features, where the sample print information is for describing a biometric print of a sample object, and the sample print features are biometric print features corresponding to the sample print information of the sample object; invoking an initial biometric print feature extraction model to process the sample print information and output a predicted fused print feature of the sample object; comparing the predicted fused print feature and the sample print feature to obtain a feature prediction error; and performing back-propagation training on the initial biometric print feature extraction model using the feature prediction error to obtain the biometric print feature extraction model.

[0182] In some embodiments of the present application, the biometric print feature extraction model is trained by: obtaining sample print information and a classification label for the sample print information, where the sample print information is for describing a biometric print of a sample object and the classification label is for identifying the sample object; invoking an initial biometric print feature extraction model to process the sample print information and output a predicted fused print feature for the sample object; performing a label classification process on the predicted fused print feature to obtain a predicted label for the sample object; comparing the predicted label with a classification label to obtain a label prediction error; and performing back-propagation training on the initial biometric print feature extraction model using the label prediction error to obtain the biometric print feature extraction model.

[0183] It should be noted that the device provided in the above embodiment is described only as an example of dividing each of the above functional modules to realize its functions, but in actual application, according to actual needs, the above functions may be assigned to different functional modules for execution, that is, the content configuration of the device may be divided into different functional modules to execute all or part of the functions described above.

[0184] Regarding the device in the above embodiment, the specific manner in which each module executes its operation is described in detail in the embodiment relating to the method, and the technical effects of executing each module's operation are the same as the technical effects in the embodiment relating to the method, so detailed explanations are omitted here.

[0185] In an embodiment of the present application, there is further provided a computer device including a processor and a memory, wherein a computer program is stored in the memory, and the processor executes the computer program in the memory to realize the biometric fingerprint information feature extraction method provided in each of the above method embodiments.

[0186] In some embodiments, the computer device is a server. Illustratively, Figure 17 is a block diagram of a server configuration provided in one exemplary embodiment of the present application.

[0187] Typically, the server 2300 includes a processor 2301 and a memory 2302 .

[0188] The processor 2301 may include one or more processing cores, for example, a 4-core processor, an 8-core processor, etc. The processor 2301 processes digital signals. Processor (DSP: Digital Signal ProcessorThe processor 2301 may be implemented in the form of at least one of a FPGA (Field-Programmable Gate Array), a Programmable Logic Array (PLA), and a memory 2302. The processor 2301 may include a main processor and a coprocessor. The main processor is a processor for processing data in a wake state and is also called 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 incorporate a graphics processing unit (GPU). The GPU is responsible for rendering and drawing content to be displayed on a display. In some embodiments, the processor 2301 may include an artificial intelligence (AI) processor for processing computational operations related to machine learning. The memory 2302 may include one or more computer-readable storage media. The computer-readable storage media may be non-transitory. The memory 2302 may include high-speed random access memory and non-volatile memory, such as one or more disk storage devices, flash memory storage devices, etc. In some embodiments, a non-transitory computer-readable storage medium in the memory 2302 stores at least one instruction that, when executed by the processor 2301, implements a biometric feature extraction method provided in a method embodiment of the present application.

[0189] In some embodiments, the server 2300 further includes 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 signal lines. Each peripheral device may be connected to the input interface 2303 and the output interface 2304 via a bus, signal lines, 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 individual chips or circuit boards. The examples of the present application are not limited thereto.

[0190] As will be appreciated by those skilled in the art, the configurations shown above are not intended to limit server 2300, and server 2300 may include more or fewer components than those shown, may combine some components, or may employ a different arrangement of components.

[0191] In an exemplary embodiment, there is further provided a chip including a programmable logic circuit and / or program instructions, which, when executed on a computing device, implements the biometric information feature extraction method described in the above aspect.

[0192] In an exemplary embodiment, there is further provided a computer program product or computer program including computer instructions, the computer instructions being stored on a computer-readable storage medium. The processor of the computing device may ,KoReading and executing the computer instructions from the computer-readable storage medium implements the biometric feature extraction method provided in each of the method embodiments above.

[0193] In an exemplary embodiment, a computer-readable storage medium storing a computer program is further provided, which, when loaded and executed by a processor, realizes the biometric fingerprint information feature extraction method provided in each of the above method embodiments.

[0194] As can be understood by those skilled in the art, all or part of the steps for realizing the above embodiments may be executed by hardware, or may be executed by instructing relevant hardware through a program. The program may be stored in a computer-readable storage medium. The storage medium mentioned above may be a read-only memory, a magnetic disk, an optical disk, etc.

[0195] Those skilled in the art should recognize that, in one or more of the above examples, the functions described in the embodiments of the present application may be implemented by hardware, software, firmware, or any combination thereof. If implemented in software, these functions may be stored on or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media. Here, communication media include any medium that facilitates the transmission of a computer program from one place to another. Storage media may be any available medium accessible by a general-purpose computer or a special-purpose computer. The above are merely some examples of the present application and are not intended to limit the present application. Various modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application should all be included within the scope of protection of the present application.

Claims

1. 1. A computer device-implemented identity recognition method, comprising: acquiring biometric fingerprint information, the biometric fingerprint information describing a biometric fingerprint of a first object; a step of acquiring an overall pattern feature and a local pattern feature by performing a feature extraction process on the biometric pattern information, the overall pattern feature being for describing an overall feature of the biometric pattern information of the first object, and the local pattern feature being for describing a local feature of the biometric pattern information of the first object; performing a self-attention mechanism process on the local pattern features to obtain updated local pattern features; performing a fusion process on the global pattern feature and the updated local pattern feature of the first object to obtain a fused pattern feature of the first object; performing identity recognition on the first object based on the fused print features of the first object; the step of acquiring overall pattern features and local pattern features by performing feature extraction processing on the biometric pattern information, obtaining texture hidden layer features of the first object by performing hidden layer feature extraction on the biometric print information, the texture hidden layer features being for describing a hidden layer representation of the biometric print information; a step of obtaining the overall pattern feature of the first object by performing feature extraction on the texture hidden layer feature using a first feature extraction method, the first feature extraction method being a feature extraction method that reduces feature dimensions using a residual convolution method; and acquiring the local pattern features of the first object by performing feature extraction on the texture hidden layer features using a second feature extraction method, wherein the second feature extraction method is a feature extraction method that expands a feature receptive field using an atlas convolution method. Identity recognition methods.

2. the step of obtaining the local pattern feature of the first object by performing feature extraction on the texture hidden layer feature using the second feature extraction method, processing the texture hidden layer features to obtain n local sub-features of the first object using n second feature extraction schemes (n being a positive integer), respectively, where the n second feature extraction schemes have different feature receptive fields; and obtaining the local print feature of the first object by combining the n local sub-features. The method of claim 1 .

3. The self-attention mechanism process includes an activation process and a regularization process; The step of acquiring updated local pattern features by performing the self-attention mechanism processing on the local pattern features includes: performing the activation process on the local pattern features to obtain second activation features; performing the regularization process on the local pattern features to obtain second regularized features; and multiplying the second activation feature by the second regularization feature to obtain the updated local print feature of the first object.

3. The method of claim 1 or 2 for recognizing an identity.

4. further comprising a step of performing preprocessing on the biometric pattern information to obtain updated biometric pattern information; The pretreatment method is as follows: performing an interpolation process on the biometric pattern information; and cropping a region of interest of the biometric fingerprint information; the step of acquiring overall pattern features and local pattern features by performing feature extraction processing on the biometric pattern information, performing a feature extraction process on the updated biometric pattern information to obtain a global pattern feature and a local pattern feature; 3. The method of claim 1 or 2 for recognizing an identity.

5. The step of performing identity recognition on the first object based on the fused print feature of the first object includes: obtaining an identity determination result for the first object by comparing the fused print feature of the first object with a sample print feature, the sample print feature being for describing a biometric print information feature of a target object; 3. The method of claim 1 or 2 for recognizing an identity.

6. The step of obtaining an identity determination result of the first object by comparing the fused print feature of the first object with the sample print feature includes: calculating a cosine similarity between the fused print feature and the sample print feature; If the cosine similarity satisfies an identity judgment condition, obtaining the identity judgment result of the first object being the target object; if the cosine similarity does not satisfy the identity determination condition, obtaining the identity determination result of the first object as not being the target object. The method of claim 5 .

7. A method for extracting features of biometric fingerprint information, which is executed by a biometric fingerprint feature extraction model of a computer device, the biometric fingerprint feature extraction model including a feature extraction network, a self-attention network, and a feature fusion network; The method for extracting features of biometric pattern information comprises: acquiring biometric fingerprint information, the biometric fingerprint information describing a biometric fingerprint of a first object; calling the feature extraction network and performing feature extraction processing on the biometric pattern information to obtain global pattern features and local pattern features, the global pattern features being for describing global features of the biometric pattern information of the first object, and the local pattern features being for describing local features of the biometric pattern information of the first object; Invoking the self-attention network to perform self-attention processing on the local pattern features to obtain updated local pattern features; and calling the feature fusion network to perform a fusion process on the global pattern feature and the updated local pattern feature of the first object, thereby obtaining a fused pattern feature of the first object; the feature extraction network includes a first residual convolution block, a second residual convolution block, and an atlas convolution block; the step of acquiring global and local pattern features by calling the feature extraction network and performing feature extraction processing on the biometric pattern information, Invoking the first residual convolution block to perform hidden layer feature extraction on the biometric print information to obtain texture hidden layer features of the first object, the texture hidden layer features being for describing a hidden layer representation of the biometric print information; calling the second residual convolution block to perform feature extraction on the texture hidden layer features using a first feature extraction method to obtain the overall pattern feature of the first object, wherein the first feature extraction method is a feature extraction method that reduces feature dimensions using a residual convolution method; and calling the atlas convolution block to perform feature extraction on the texture hidden layer features using a second feature extraction scheme to obtain the local pattern features of the first object, wherein the second feature extraction scheme is a feature extraction scheme that expands a feature receptive field using an atlas convolution scheme. A method for extracting features from biometric fingerprint information.

8. The atlas convolution block includes n atlas convolution layers (n is a positive integer), The step of obtaining the local pattern features of the first object by calling the atlas convolution block to perform feature extraction on the texture hidden layer features using a second feature extraction scheme, includes: Invoking the n atlas convolutional layers to process the texture hidden layer features and obtain n local sub-features of the first object using n second feature extraction methods, respectively, where the n second feature extraction methods have different feature receptive fields, and the feature receptive fields are determined based on a convolution kernel scale of the atlas convolutional layer; and obtaining the local print feature of the first object by combining the n local sub-features. The method for extracting features of biometric pattern information according to claim 7.

9. The biometric fingerprint feature extraction model is acquiring sample print information and sample print features, the sample print information describing a biometric print of a sample object, and the sample print features being biometric print features corresponding to the sample print information of the sample object; invoking an initial biometric fingerprint feature extraction model to process the sample fingerprint information and output a predicted fused biometric fingerprint feature of the sample object; obtaining a feature prediction error by comparing the predicted fused print feature with a sample print feature; and performing back-propagation training on the initial biometric fingerprint feature extraction model using the feature prediction error to obtain the biometric fingerprint feature extraction model.

9. The method for extracting features of biometric pattern information according to claim 7 or 8.

10. The biometric fingerprint feature extraction model is obtaining sample print information and a classification label for the sample print information, the sample print information for describing a biometric print of a sample object, and the classification label for identifying the sample object; invoking an initial biometric fingerprint feature extraction model to process the sample fingerprint information and output a predicted fused biometric fingerprint feature of the sample object; performing a label classification process on the predicted fused print features to obtain a predicted label for the sample object; obtaining a label prediction error by comparing the predicted label with the classification label; and performing back-propagation training on the initial biometric fingerprint feature extraction model using the label prediction error to obtain the biometric fingerprint feature extraction model.

9. The method for extracting features of biometric pattern information according to claim 7 or 8.

11. an acquisition module for acquiring biometric fingerprint information, the biometric fingerprint information describing a biometric fingerprint of a first object; an extraction module that acquires global pattern features and local pattern features by performing feature extraction processing on the biometric pattern information, wherein the global pattern features are for describing global features of the biometric pattern information of the first object, and the local pattern features are for describing local features of the biometric pattern information of the first object; a processing module for performing self-attention processing on the local pattern features to obtain updated local pattern features; a fusion module that performs a fusion process on the global pattern feature and the updated local pattern feature of the first object to obtain a fused pattern feature of the first object; a determination module for performing identity recognition on the first object based on the fusion print feature of the first object; The extraction module: a first extraction module for performing hidden layer feature extraction on the biometric fingerprint information to obtain texture hidden layer features of the first object, the texture hidden layer features being for describing a hidden layer representation of the biometric fingerprint information; a second extraction module for obtaining the overall pattern feature of the first object by performing feature extraction on the texture hidden layer feature using a first feature extraction method, the first feature extraction method being a feature extraction method for reducing feature dimensions using a residual convolution method; a third extraction module for obtaining the local pattern features of the first object by performing feature extraction on the texture hidden layer features using a second feature extraction method, the second feature extraction method being a feature extraction method for expanding a feature receptive field using an atlas convolution method; Identity recognition device.

12. A biometric fingerprint information feature extraction device including a biometric fingerprint feature extraction model, the biometric fingerprint feature extraction model including a feature extraction network, a self-attention network, and a feature fusion network; The biometric pattern information feature extraction device comprises: an acquisition module for acquiring biometric fingerprint information, the biometric fingerprint information describing a biometric fingerprint of a first object; an extraction module that calls the feature extraction network and performs feature extraction processing on the biometric pattern information to acquire global pattern features and local pattern features, the global pattern features being for describing global features of the biometric pattern information of the first object, and the local pattern features being for describing local features of the biometric pattern information of the first object; a processing module that calls the self-attention network and performs self-attention processing on the local pattern features to obtain updated local pattern features; a fusion module that calls the feature fusion network and performs a fusion process on the global pattern feature and the updated local pattern feature of the first object to obtain a fused pattern feature of the first object, the feature extraction network includes a first residual convolution block, a second residual convolution block, and an atlas convolution block; The extraction module: a first extraction module that calls the first residual convolution block to perform hidden layer feature extraction on the biometric print information to obtain texture hidden layer features of the first object, the texture hidden layer features being for describing a hidden layer representation of the biometric print information; a second extraction module that calls the second residual convolution block to perform feature extraction on the texture hidden layer features using a first feature extraction method to obtain the overall pattern feature of the first object, wherein the first feature extraction method is a feature extraction method that reduces feature dimensions using a residual convolution method; a third extraction module that calls the atlas convolution block to perform feature extraction on the texture hidden layer features using a second feature extraction method to obtain the local pattern features of the first object, and the second feature extraction method is a feature extraction method that expands a feature receptive field using an atlas convolution method; A device for extracting features from biometric fingerprint information.

13. 10. A computing device comprising a processor and a memory, wherein the memory stores at least one program, and the processor executes the at least one program in the memory to implement the identity recognition method of claim 1.

14. A computer device comprising a processor and a memory, wherein at least one program is stored in the memory, and the processor executes the at least one program in the memory to realize the method for extracting features of biometric pattern information described in claim 7.

15. A computer program product that causes a computer to perform the identity recognition method of claim 1.

16. A computer program that causes a computer to execute the method for extracting features of biometric pattern information described in claim 7.

Images