Fingerprint feature recognition method, device and equipment and readable storage medium
Through the combination of heat map generation network, local feature network and feature conversion network, the problem of low semi-fingerprint recognition accuracy in traditional technology is solved, and high-precision and robust fingerprint feature recognition is achieved, which improves the reliability of the system and User experience.
Patent Information
- Application Number
- CN202411584813.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-07
- Publication Date
- 2025-01-14
AI Technical Summary
When traditional fingerprint recognition technology processes semi-fingerprints, the recognition accuracy is low, resulting in an increase in the risk of misidentification, especially due to insufficient global feature expression capabilities, the inability to effectively utilize local features, affecting the reliability and user experience of the system.
A combination of heat map generation network, local feature network and feature conversion network is used to accurately locate local feature points, efficiently extract local features and optimize feature representations, a fingerprint feature recognition model is formed to improve the recognition accuracy and robustness of half-fingerprints.
High-precision recognition of half-fingerprints is realized, the model's generalization ability of fingerprint features is enhanced, feature representation is simplified, computing complexity is reduced, and the reliability and user experience of the recognition system are improved.
Smart Images
Figure CN119314207A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of feature recognition, and more specifically, to a fingerprint feature recognition method, device, equipment and readable storage medium. Background Art
[0002] In recent years, the rapid progress of electronic devices and security verification technology has promoted the widespread application of fingerprint recognition technology in identity authentication and security systems. The popularity of this technology is mainly due to the uniqueness and high stability of fingerprints, which make them difficult to copy or forge, thus ensuring the accuracy and security of identity authentication. However, in practical applications, the accuracy of fingerprint recognition is still a crucial challenge.
[0003] In order to ensure recognition accuracy, users are usually required to provide complete fingerprint information, that is, full fingerprints, when registering fingerprints. However, in the actual verification process, due to differences in user operating habits, fingerprint collection equipment, and external factors, the system can often only capture part of the fingerprint, that is, half fingerprints. Traditional deep learning methods, especially convolutional neural network (CNN) structures, perform well when processing complete fingerprints and can accurately extract global features. However, when faced with half fingerprints, the expressive power of these global features is weak. Due to incomplete information or large differences in local features, directly comparing the global features of half fingerprints and full fingerprints of the same ID often results in low similarity scores, which increases the risk of misidentification and reduces the reliability of the fingerprint recognition system and user experience.
[0004] In order to solve this problem, an innovative fingerprint recognition method is proposed, which is not limited to traditional minutiae and has stronger local feature expression ability, thus improving the recognition accuracy. Summary of the invention
[0005] In view of this, the present application provides a fingerprint feature recognition method, apparatus, device and readable storage medium, which achieves high accuracy and strong robustness of half-fingerprint recognition by accurately locating local feature points, efficiently extracting local features and optimizing feature representation.
[0006] A fingerprint feature recognition method, comprising:
[0007] Obtain target fingerprint image;
[0008] Determine a fingerprint feature recognition model, the fingerprint feature recognition model includes a heat map generation network, a local feature network, and a feature conversion network, the heat map generation network is used to determine the local feature points in the target fingerprint image, and generate a heat map based on the local feature points, the local feature network is used to extract local detail features from the heat map, generate each detail feature map, the feature conversion network is used to convert each detail feature map into each low-dimensional feature vector, and integrate each low-dimensional feature vector into a fingerprint feature recognition result;
[0009] The target fingerprint image is input into the fingerprint feature recognition model to obtain a fingerprint feature recognition result of the target fingerprint image output by the fingerprint feature recognition model.
[0010] Optionally, determining the fingerprint feature recognition model includes:
[0011] Acquire a sample fingerprint image set, wherein each sample fingerprint image in the sample fingerprint image set is annotated with a corresponding fingerprint feature result;
[0012] The sample fingerprint image set is input into a preset initial feature recognition model, and the initial feature recognition model is iteratively trained with the goal of minimizing the local feature loss function, so as to obtain the optimal initial feature recognition model based on the local feature loss function after training, and use it as the fingerprint feature recognition model.
[0013] Optionally, the local feature loss function is:
[0014]
[0015]
[0016] Among them, emb_local is a low-dimensional feature vector with size dimension and weight N*emb_dim, The size dimension is 1*emb_dim, m is the dimension of the low-dimensional feature vector, n is the number of local feature points, and s is the fixed value for normalizing the feature vector.
[0017] Optionally, the fingerprint feature recognition model calculates the fingerprint feature recognition result of the target fingerprint image by analyzing the target fingerprint image input into the fingerprint feature recognition model as follows:
[0018]
[0019] Among them, emb_local is a low-dimensional feature vector with size dimension and weight both being N*emb_dim.
[0020] Optionally, the heat map generation network includes a first convolutional layer and a second convolutional layer;
[0021] The first convolutional layer is used to extract features of the input target fingerprint image to obtain transition fingerprint features;
[0022] The second convolutional layer is used to determine local feature points based on the intermediate features of the transition fingerprint, and generate a heat map whose number of channels is consistent with the number of the local feature points, wherein the maximum value position of the heat map is the center position of the corresponding local feature.
[0023] Optionally, the local feature network includes a feature extraction layer and a feature map construction layer;
[0024] The feature extraction layer is used to extract each of the local detail features from the thermal map;
[0025] The feature map construction layer is used to generate each detail feature map having the same number as the local feature points based on each of the local detail features.
[0026] Optionally, the feature conversion network includes a vector conversion layer and a vector integration layer;
[0027] The vector conversion layer is used to convert and generate low-dimensional feature vectors by performing feature extraction and dimensionality reduction processing on the detail feature maps;
[0028] The vector integration layer is used to integrate the low-dimensional feature vectors into a fingerprint feature recognition result.
[0029] A fingerprint feature recognition device, comprising:
[0030] An image acquisition unit, used to acquire a target fingerprint image;
[0031] A model determination unit, used to determine a fingerprint feature recognition model, wherein the fingerprint feature recognition model includes a heat map generation network, a local feature network, and a feature conversion network. The heat map generation network is used to determine local feature points in the target fingerprint image and generate a heat map based on the local feature points. The local feature network is used to extract local detail features from the heat map to generate detail feature maps. The feature conversion network is used to convert the detail feature maps into low-dimensional feature vectors, and integrate the low-dimensional feature vectors into a fingerprint feature recognition result.
[0032] The feature recognition unit is used to input the target fingerprint image into the fingerprint feature recognition model to obtain the fingerprint feature recognition result of the target fingerprint image output by the fingerprint feature recognition model.
[0033] A fingerprint feature recognition device, comprising a memory and a processor;
[0034] The memory is used to store programs;
[0035] The processor is used to execute the program to implement the various steps of the fingerprint feature recognition method as described above.
[0036] A readable storage medium stores a computer program, which, when executed by a processor, implements the various steps of the fingerprint feature recognition method as described above.
[0037] It can be seen from the above technical solutions that the embodiments of the present application provide a fingerprint feature recognition method, device, equipment and readable storage medium, by inputting the target fingerprint image into the fingerprint feature recognition model, to obtain the fingerprint feature recognition result of the target fingerprint image output by the fingerprint feature recognition model, wherein the fingerprint feature recognition model includes a heat map generation network, a local feature network, and a feature conversion network, the heat map generation network is used to determine the local feature points in the target fingerprint image, and generate a heat map based on the local feature points, the local feature network is used to extract local detail features from the heat map, and generate each detail feature map, the feature conversion network is used to convert each detail feature map into each low-dimensional feature vector, and integrate each low-dimensional feature vector into a fingerprint feature recognition result.
[0038] The heat map generation network of the fingerprint feature recognition model can accurately locate the local feature points in the target fingerprint image and generate a heat map based on these feature points. The local feature network uses the information in the heat map to extract local detail features from around each local feature point and generate a corresponding detail feature map, which not only enhances the sensitivity to local features, but also improves the utilization rate of half-fingerprint information, thereby effectively improving the recognition accuracy. The feature conversion network converts each detail feature map into a low-dimensional feature vector and integrates these vectors to form the final fingerprint feature recognition result, which not only simplifies the feature representation and reduces the computational complexity, but also enhances the model's generalization ability for fingerprint features, enabling the model to more accurately identify fingerprint information under different conditions. Therefore, the present application achieves high precision and strong robustness in half-fingerprint recognition by accurately locating local feature points, efficiently extracting local features, and optimizing feature representation. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.
[0040] Figure 1 A flow chart of a fingerprint feature recognition method disclosed in this application;
[0041] Figure 2 A schematic diagram of the structure of a fingerprint feature recognition model disclosed in this application;
[0042] Figure 3 A schematic diagram of the structure of another fingerprint feature recognition model disclosed in this application;
[0043] Figure 4 This is a structural block diagram of a fingerprint feature recognition device disclosed in this application;
[0044] Figure 5 This is a hardware structure block diagram of a fingerprint feature recognition device disclosed in this application. DETAILED DESCRIPTION
[0045] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0046] Next, the present application scheme is introduced. The present application proposes the following technical scheme, please see below for details.
[0047] Figure 1 This is a flow chart of a fingerprint feature recognition method disclosed in an embodiment of the present application. The fingerprint feature recognition method can be applied to a fingerprint feature recognition device, such as Figure 1 As shown, the method may include:
[0048] Step S1, obtaining a target fingerprint image.
[0049] Specifically, the acquisition of the target fingerprint image is usually completed through a high-precision fingerprint acquisition device to ensure that the image quality meets the needs of subsequent processing. The fingerprint acquisition device can be, for example, a mobile phone, an access control system, a security device, etc. In actual applications, when registering a fingerprint, the user is usually required to provide a complete fingerprint, also called a full fingerprint, to ensure a high degree of detail and recognition accuracy. However, in the actual use and verification process, due to various reasons, such as user operating habits, differences in fingerprint acquisition equipment, or other external factors, only a partial fingerprint, i.e., a half fingerprint, is often obtained. The target fingerprint image can be a full fingerprint or a half fingerprint. When the detected target fingerprint image is a half fingerprint, the superiority of the present application over the prior art can be better reflected.
[0050] In order to improve recognition accuracy, it may be necessary to pre-process the image, such as denoising, contrast enhancement, etc., to highlight the detailed features of the fingerprint.
[0051] Since the fingerprint collection process may be interfered by various noises, such as equipment noise, environmental noise, etc., these noises will blur the detailed features of the fingerprint and affect the accuracy of recognition. Therefore, the denoising process uses filters or other algorithms to reduce or eliminate the noise in the image to make the fingerprint texture clearer.
[0052] The fine details in fingerprint images, such as the subtle differences in ridges and valleys, are crucial for recognition. However, due to different acquisition conditions, the contrast of fingerprint images may vary. By enhancing the contrast, the differences between the ridges and valleys of the fingerprint can be made more obvious, thereby improving the accuracy of recognition. Contrast enhancement is usually achieved by adjusting the brightness and contrast parameters of the image, or using more advanced image processing algorithms such as adaptive histogram equalization.
[0053] In addition, preprocessing may also include other steps such as image cropping, resizing, rotation correction, etc. to ensure the consistency and standardization of the image, which helps to reduce recognition errors caused by differences in image size and orientation.
[0054] Step S2: determine the fingerprint feature recognition model.
[0055] Specifically, Figure 2 As shown, the fingerprint feature recognition model includes a heat map generation network A, a local feature network B, and a feature conversion network C. The heat map generation network is used to determine the local feature points in the target fingerprint image and generate a heat map based on the local feature points. The local feature network is used to extract local detail features from the heat map and generate each detail feature map. The feature conversion network is used to convert each detail feature map into each low-dimensional feature vector, and integrate each low-dimensional feature vector into a fingerprint feature recognition result.
[0056] Specifically, the heat map generation network is responsible for accurately locating the local feature points in the target fingerprint image. These local feature points are usually key details in the fingerprint image, such as the endpoints, bifurcation points, and intersections of the ridges. Based on the location information of these local feature points, the heat map generation network can generate a heat map with the same number of channels as the local feature points, which shows the distribution of local feature points in the fingerprint image in an intuitive way.
[0057] Next, the local feature network uses the information in the heat map to extract local detail features from around each feature point. These detail features are usually fingerprint texture and shape information contained in a small area centered on the feature point. The local feature network extracts features from each detail through an image processing algorithm and generates corresponding detail feature maps. These feature maps not only retain the original information of the details, but also enhance the model's sensitivity to local fingerprint features through feature encoding.
[0058] Finally, the feature conversion network converts each detail feature map into a low-dimensional feature vector. This process is the key to the model's efficient feature representation. By converting high-dimensional detail feature maps into low-dimensional feature vectors, the feature conversion network not only simplifies feature representation and reduces computational complexity, but also improves the model's generalization ability for fingerprint features. These low-dimensional feature vectors are not only easy to store and transmit, but also easy to quickly match with fingerprint features in the database. Finally, the feature conversion network integrates all detail feature vectors into a single fingerprint feature recognition result, which can accurately reflect the uniqueness and stability of the fingerprint.
[0059] It is understandable that in order to further shorten the detection time and improve the detection efficiency, the present application can pre-train a preset initial feature recognition model based on the sample fingerprint image set. Each sample fingerprint image in the sample fingerprint image set is annotated with a corresponding fingerprint feature result. The initial feature recognition model after training is determined as the fingerprint feature recognition model, that is, when the target fingerprint image is first acquired and fingerprint feature recognition is performed, the corresponding fingerprint feature recognition model can be determined and saved. When fingerprint feature recognition is performed on the target fingerprint image later, the fingerprint feature recognition model can be directly used without the need to determine it again. Among them, the initial feature recognition model also includes a heat map generation network, a local feature network, and a feature conversion network, and it is iteratively trained with the goal of minimizing the local feature loss function. During the training process, the various parameters of each network structure are continuously adjusted and corrected.
[0060] Step S3: input the target fingerprint image into the fingerprint feature recognition model to obtain the fingerprint feature recognition result of the target fingerprint image output by the fingerprint feature recognition model.
[0061] Specifically, the preprocessed target fingerprint image is input into the fingerprint feature recognition model, and the fingerprint feature recognition result of the target fingerprint image is finally output. Specifically, when the target fingerprint image is input into the model, it first enters the heat map generation network, which uses a deep learning algorithm to analyze the fingerprint image pixel by pixel to determine the local feature points therein. The local feature network uses the information in the heat map to extract features from the local area around each feature point, extracts local detail features from the heat map, and generates each detail feature map. The feature conversion network converts each detail feature map into a low-dimensional feature vector, and integrates these vectors to form the final fingerprint feature recognition result.
[0062] The fingerprint feature recognition model calculates the fingerprint feature recognition result of the target fingerprint image by analyzing the target fingerprint image input into the fingerprint feature recognition model as follows:
[0063]
[0064] Among them, emb_local is a low-dimensional feature vector with size dimension and weight both being N*emb_dim.
[0065] It can be seen from the above technical solutions that the embodiments of the present application provide a fingerprint feature recognition method, device, equipment and readable storage medium, by inputting the target fingerprint image into the fingerprint feature recognition model, to obtain the fingerprint feature recognition result of the target fingerprint image output by the fingerprint feature recognition model, wherein the fingerprint feature recognition model includes a heat map generation network, a local feature network, and a feature conversion network, the heat map generation network is used to determine the local feature points in the target fingerprint image, and generate a heat map based on the local feature points, the local feature network is used to extract local detail features from the heat map, and generate each detail feature map, the feature conversion network is used to convert each detail feature map into each low-dimensional feature vector, and integrate each low-dimensional feature vector into a fingerprint feature recognition result.
[0066] The heat map generation network of the fingerprint feature recognition model can accurately locate the local feature points in the target fingerprint image and generate a heat map based on these feature points. The local feature network uses the information in the heat map to extract local detail features from around each local feature point and generate a corresponding detail feature map, which not only enhances the sensitivity to local features, but also improves the utilization rate of half-fingerprint information, thereby effectively improving the recognition accuracy. The feature conversion network converts each detail feature map into a low-dimensional feature vector and integrates these vectors to form the final fingerprint feature recognition result, which not only simplifies the feature representation and reduces the computational complexity, but also enhances the model's generalization ability for fingerprint features, enabling the model to more accurately identify fingerprint information under different conditions. Therefore, the present application achieves high precision and strong robustness in half-fingerprint recognition by accurately locating local feature points, efficiently extracting local features, and optimizing feature representation.
[0067] The above embodiment describes the process of the fingerprint feature recognition method in the present application. The following will describe the process of determining the fingerprint feature recognition model in the present application.
[0068] In some embodiments of the present application, determining the fingerprint feature recognition model may include:
[0069] Step S21, obtaining a sample fingerprint image set.
[0070] Specifically, the quality of the sample fingerprint image set directly determines the performance of the trained model. Specifically, the sample fingerprint image set should contain a large number of fingerprint images of different individuals and under different conditions, and each sample fingerprint image should be annotated with the corresponding fingerprint feature results. These annotated results can be manually annotated or obtained through other high-precision methods.
[0071] Step S22: input the sample fingerprint image set into a preset initial feature recognition model, iteratively train the initial feature recognition model with the goal of minimizing the local feature loss function, obtain the optimal initial feature recognition model based on the local feature loss function after training, and use it as the fingerprint feature recognition model.
[0072] Specifically, the sample fingerprint image set is input into the preset initial feature recognition model and iterative training is performed. The training goal is to minimize the local feature loss function to improve the model's sensitivity to the local features of the fingerprint and the recognition accuracy.
[0073] Wherein, the local feature loss function is:
[0074]
[0075]
[0076] Among them, emb_local is a low-dimensional feature vector with size dimension and weight N*emb_dim, The size dimension is 1*emb_dim, m is the dimension of the low-dimensional feature vector, n is the number of local feature points, and s is the fixed value for normalizing the feature vector.
[0077] During the training process, the model continuously adjusts the network parameters to minimize the local feature loss function. This means that the model maximizes (or minimizes) the cosine similarity between the low-dimensional feature vector of each local feature point and the corresponding reference vector (or fingerprint feature in the database), thereby improving the recognition accuracy of the model. After multiple iterations of training, when the local feature loss function converges to a smaller value, it can be considered that the model has been trained and the optimal initial feature recognition model based on the local feature loss function has been obtained. At this point, the model can be used as a fingerprint feature recognition model for subsequent fingerprint recognition tasks.
[0078] In some embodiments of the present application, the structure and function of the fingerprint feature recognition model of the present application are specifically introduced, and the fingerprint feature recognition model includes a heat map generation network A, a local feature network B, and a feature conversion network C. Figure 3 , the heat map generation network A, local feature network B, and feature conversion network C are introduced in detail in turn.
[0079] Heat map generation network A:
[0080] Optionally, the heat map generation network A may include a first convolutional layer A1 and a second convolutional layer A2.
[0081] The first convolutional layer is used to extract features from the input target fingerprint image to obtain transition fingerprint features;
[0082] The second convolutional layer is used to determine local feature points based on the intermediate features of the transition fingerprint, and generate a heat map whose number of channels is consistent with the number of the local feature points, wherein the maximum value position of the heat map is the center position of the corresponding local feature.
[0083] Specifically, the first convolutional layer is the entrance to the heat map generation network, and its main task is to perform preliminary feature extraction on the input fingerprint image. By applying convolution operations, this layer can capture the basic texture and shape information in the fingerprint image. After being processed by the first convolutional layer, the output shape size is 512*14*14.
[0084] The second convolutional layer focuses on determining the local feature points in the fingerprint image. These feature points are usually key points in the fingerprint image, such as the end points and bifurcation points of the ridges, which play a key role in the unique identification of the fingerprint. After the second convolutional layer, a heat map containing the position and intensity information of all key feature points in the fingerprint image can be obtained. The number of channels in the heat map is consistent with the number of local feature points. Each channel corresponds to a feature point type, and the maximum value position in the heat map represents the center position of the corresponding feature point. Through the representation of the heat map, the distribution and intensity of each feature point in the fingerprint image can be intuitively seen.
[0085] Local feature network B:
[0086] Optionally, the local feature network B may include a feature extraction layer B1 and a feature map construction layer B2.
[0087] The feature extraction layer is used to extract each of the local detail features from the thermal map;
[0088] The feature map construction layer is used to generate each detail feature map having the same number as the local feature points based on each of the local detail features.
[0089] Specifically, the feature extraction layer is responsible for extracting each local detail feature from the heat map. These features usually contain detailed texture and shape information of key points in the fingerprint image, which is an important basis for distinguishing different fingerprints. Specifically, it receives the heat map as input, captures and outputs the feature vector of each local detail through convolution operations, pooling and other algorithms. Subsequently, the feature map construction layer further constructs each detail feature map consistent with the number of local feature points on the basis of the feature extraction layer, and the shape and size of the detail feature map is 512*3*3. These feature maps not only contain detailed information of local detail features, but also significantly enhance the model's sensitivity and recognition ability to fingerprint features through spatial arrangement and encoding. Finally, the local feature network outputs a series of detail feature maps, which will be used as input to the subsequent feature conversion network to generate low-dimensional feature vectors, thereby completing the extraction and representation of fingerprint features.
[0090] Feature conversion network C:
[0091] Optionally, the feature conversion network C may include a vector conversion layer C1 and a vector integration layer C2.
[0092] The vector conversion layer is used to convert and generate low-dimensional feature vectors by performing feature extraction and dimensionality reduction processing on the detail feature maps;
[0093] The vector integration layer is used to integrate the low-dimensional feature vectors into a fingerprint feature recognition result.
[0094] Specifically, the vector conversion layer is responsible for receiving the detailed feature maps from the local feature network, and converting these high-dimensional feature maps into low-dimensional feature vectors through further feature extraction and dimensionality reduction processing. This process aims to retain the key information of fingerprint features while reducing the dimension of the data and improving computational efficiency. Subsequently, the vector integration layer integrates the various low-dimensional feature vectors and generates the final fingerprint feature recognition results by utilizing their complementarity and correlation. This integration process not only enhances the model's comprehensive understanding and recognition capabilities of fingerprint features, but also provides a more accurate and reliable basis for subsequent fingerprint recognition and identity authentication.
[0095] The fingerprint feature recognition device provided in the embodiment of the present application is described below. The fingerprint feature recognition device described below and the fingerprint feature recognition method described above can be referenced to each other.
[0096] See also Figure 4 , Figure 4 This is a structural block diagram of a fingerprint feature recognition device disclosed in an embodiment of the present application.
[0097] like Figure 4 As shown, the fingerprint feature recognition device may include:
[0098] An image acquisition unit 110 is used to acquire a target fingerprint image;
[0099] The model determination unit 120 is used to determine a fingerprint feature recognition model, wherein the fingerprint feature recognition model includes a heat map generation network, a local feature network, and a feature conversion network. The heat map generation network is used to determine local feature points in the target fingerprint image and generate a heat map based on the local feature points. The local feature network is used to extract local detail features from the heat map to generate detail feature maps. The feature conversion network is used to convert the detail feature maps into low-dimensional feature vectors, and integrate the low-dimensional feature vectors into a fingerprint feature recognition result.
[0100] The feature recognition unit 130 is used to input the target fingerprint image into the fingerprint feature recognition model to obtain the fingerprint feature recognition result of the target fingerprint image output by the fingerprint feature recognition model.
[0101] It can be seen from the above technical solutions that the embodiments of the present application provide a fingerprint feature recognition method, device, equipment and readable storage medium, by inputting the target fingerprint image into the fingerprint feature recognition model, to obtain the fingerprint feature recognition result of the target fingerprint image output by the fingerprint feature recognition model, wherein the fingerprint feature recognition model includes a heat map generation network, a local feature network, and a feature conversion network, the heat map generation network is used to determine the local feature points in the target fingerprint image, and generate a heat map based on the local feature points, the local feature network is used to extract local detail features from the heat map, and generate each detail feature map, the feature conversion network is used to convert each detail feature map into each low-dimensional feature vector, and integrate each low-dimensional feature vector into a fingerprint feature recognition result.
[0102] The heat map generation network of the fingerprint feature recognition model can accurately locate the local feature points in the target fingerprint image and generate a heat map based on these feature points. The local feature network uses the information in the heat map to extract local detail features from around each local feature point and generate a corresponding detail feature map, which not only enhances the sensitivity to local features, but also improves the utilization rate of half-fingerprint information, thereby effectively improving the recognition accuracy. The feature conversion network converts each detail feature map into a low-dimensional feature vector and integrates these vectors to form the final fingerprint feature recognition result, which not only simplifies the feature representation and reduces the computational complexity, but also enhances the model's generalization ability for fingerprint features, enabling the model to more accurately identify fingerprint information under different conditions. Therefore, the present application achieves high precision and strong robustness in half-fingerprint recognition by accurately locating local feature points, efficiently extracting local features, and optimizing feature representation.
[0103] Optionally, the model determination module 120 may include:
[0104] A sample acquisition unit, used to acquire a sample fingerprint image set, each sample fingerprint image in the sample fingerprint image set is marked with a corresponding fingerprint feature result;
[0105] The sample training unit is used to input the sample fingerprint image set into a preset initial feature recognition model, iteratively train the initial feature recognition model with the goal of minimizing the local feature loss function, obtain the optimal initial feature recognition model based on the local feature loss function after training, and use it as the fingerprint feature recognition model.
[0106] Optionally, the local feature loss function is:
[0107]
[0108]
[0109] Among them, emb_local is a low-dimensional feature vector with size dimension and weight N*emb_dim, The size dimension is 1*emb_dim, m is the dimension of the low-dimensional feature vector, n is the number of local feature points, and s is the fixed value for normalizing the feature vector.
[0110] Optionally, the fingerprint feature recognition model calculates the fingerprint feature recognition result of the target fingerprint image by analyzing the target fingerprint image input into the fingerprint feature recognition model as follows:
[0111]
[0112] Among them, emb_local is a low-dimensional feature vector with size dimension and weight both being N*emb_dim.
[0113] Optionally, the heat map generation network includes a first convolutional layer and a second convolutional layer;
[0114] The first convolutional layer is used to extract features from the input target fingerprint image to obtain transition fingerprint features;
[0115] The second convolutional layer is used to determine local feature points based on the intermediate features of the transition fingerprint, and generate a heat map whose number of channels is consistent with the number of the local feature points, wherein the maximum value position of the heat map is the center position of the corresponding local feature.
[0116] Optionally, the local feature network includes a feature extraction layer and a feature map construction layer;
[0117] The feature extraction layer is used to extract each of the local detail features from the thermal map;
[0118] The feature map construction layer is used to generate each detail feature map having the same number as the local feature points based on each of the local detail features.
[0119] Optionally, the feature conversion network includes a vector conversion layer and a vector integration layer;
[0120] The vector conversion layer is used to convert and generate low-dimensional feature vectors by performing feature extraction and dimensionality reduction processing on the detail feature maps;
[0121] The vector integration layer is used to integrate the low-dimensional feature vectors into a fingerprint feature recognition result.
[0122] The fingerprint feature recognition device provided in the embodiment of the present application can be applied to fingerprint feature recognition equipment. Figure 5 The hardware structure diagram of the fingerprint feature recognition device is shown in FIG. Figure 5,The hardware structure of the fingerprint feature recognition device may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;
[0123] In the embodiment of the present application, the number of the processor 1, the communication interface 2, the memory 3, and the communication bus 4 is at least one, and the processor 1, the communication interface 2, and the memory 3 communicate with each other through the communication bus 4;
[0124] The processor 1 may be a central processing unit CPU, or an application-specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention, etc.;
[0125] The memory 3 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), etc., such as at least one disk memory;
[0126] The memory stores a program, and the processor can call the program stored in the memory, wherein the program is used to:
[0127] Obtain target fingerprint image;
[0128] Determine a fingerprint feature recognition model, the fingerprint feature recognition model includes a heat map generation network, a local feature network, and a feature conversion network, the heat map generation network is used to determine the local feature points in the target fingerprint image, and generate a heat map based on the local feature points, the local feature network is used to extract local detail features from the heat map, generate each detail feature map, the feature conversion network is used to convert each detail feature map into each low-dimensional feature vector, and integrate each low-dimensional feature vector into a fingerprint feature recognition result;
[0129] The target fingerprint image is input into the fingerprint feature recognition model to obtain a fingerprint feature recognition result of the target fingerprint image output by the fingerprint feature recognition model.
[0130] Optionally, the detailed functions and extended functions of the program may refer to the above description.
[0131] The embodiment of the present application further provides a readable storage medium, which may store a program suitable for execution by a processor, wherein the program is used to:
[0132] Obtain target fingerprint image;
[0133] Determine a fingerprint feature recognition model, the fingerprint feature recognition model includes a heat map generation network, a local feature network, and a feature conversion network, the heat map generation network is used to determine the local feature points in the target fingerprint image, and generate a heat map based on the local feature points, the local feature network is used to extract local detail features from the heat map, generate each detail feature map, the feature conversion network is used to convert each detail feature map into each low-dimensional feature vector, and integrate each low-dimensional feature vector into a fingerprint feature recognition result;
[0134] The target fingerprint image is input into the fingerprint feature recognition model to obtain a fingerprint feature recognition result of the target fingerprint image output by the fingerprint feature recognition model.
[0135] Optionally, the detailed functions and extended functions of the program may refer to the above description.
[0136] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
[0137] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.
[0138] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A fingerprint feature recognition method, characterized in that: include: Obtain target fingerprint image; Determine a fingerprint feature recognition model, the fingerprint feature recognition model includes a heat map generation network, a local feature network, and a feature conversion network, the heat map generation network is used to determine the local feature points in the target fingerprint image, and generate a heat map based on the local feature points, the local feature network is used to extract local detail features from the heat map, generate each detail feature map, the feature conversion network is used to convert each detail feature map into each low-dimensional feature vector, and integrate each low-dimensional feature vector into a fingerprint feature recognition result; The target fingerprint image is input into the fingerprint feature recognition model to obtain a fingerprint feature recognition result of the target fingerprint image output by the fingerprint feature recognition model.
2. The method according to claim 1, characterized in that The step of determining the fingerprint feature recognition model comprises: Acquire a sample fingerprint image set, wherein each sample fingerprint image in the sample fingerprint image set is annotated with a corresponding fingerprint feature result; The sample fingerprint image set is input into a preset initial feature recognition model, and the initial feature recognition model is iteratively trained with the goal of minimizing the local feature loss function, so as to obtain the optimal initial feature recognition model based on the local feature loss function after training, and use it as the fingerprint feature recognition model.
3. The method according to claim 2, characterized in that The local feature loss function is: Among them, emb_local is a low-dimensional feature vector with size dimension and weight N*emb_dim, The size dimension is 1*emb_dim, m is the dimension of the low-dimensional feature vector, n is the number of local feature points, and s is the fixed value for normalizing the feature vector.
4. The method according to claim 1, characterized in that: The fingerprint feature recognition model calculates the fingerprint feature recognition result of the target fingerprint image by analyzing the target fingerprint image according to the input target fingerprint image into the fingerprint feature recognition model: Among them, emb_local is a low-dimensional feature vector with size dimension and weight both being N*emb_dim.
5. The method according to claim 1, characterized in that The heat map generation network includes a first convolutional layer and a second convolutional layer; The first convolutional layer is used to extract features of the input target fingerprint image to obtain transition fingerprint features; The second convolutional layer is used to determine local feature points based on the intermediate features of the transition fingerprint, and generate a heat map whose number of channels is consistent with the number of the local feature points, wherein the maximum value position of the heat map is the center position of the corresponding local feature.
6. The method according to claim 1, characterized in that The local feature network includes a feature extraction layer and a feature map construction layer; The feature extraction layer is used to extract each of the local detail features from the thermal map; The feature map construction layer is used to generate each detail feature map having the same number as the local feature points based on each of the local detail features.
7. The method according to claim 1, characterized in that The feature conversion network includes a vector conversion layer and a vector integration layer; The vector conversion layer is used to convert and generate low-dimensional feature vectors by performing feature extraction and dimensionality reduction processing on the detail feature maps; The vector integration layer is used to integrate the low-dimensional feature vectors into a fingerprint feature recognition result.
8. A fingerprint feature recognition device, characterized in that: include: An image acquisition unit, used to acquire a target fingerprint image; A model determination unit, used to determine a fingerprint feature recognition model, wherein the fingerprint feature recognition model includes a heat map generation network, a local feature network, and a feature conversion network. The heat map generation network is used to determine local feature points in the target fingerprint image and generate a heat map based on the local feature points. The local feature network is used to extract local detail features from the heat map to generate detail feature maps. The feature conversion network is used to convert the detail feature maps into low-dimensional feature vectors, and integrate the low-dimensional feature vectors into a fingerprint feature recognition result. The feature recognition unit is used to input the target fingerprint image into the fingerprint feature recognition model to obtain the fingerprint feature recognition result of the target fingerprint image output by the fingerprint feature recognition model.
9. A fingerprint feature recognition device, characterized in that: including memory and processor; The memory is used to store programs; The processor is used to execute the program to implement each step of the fingerprint feature recognition method as described in any one of claims 1 to 7.
10. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the fingerprint feature recognition method according to any one of claims 1 to 7 is implemented.