Fingerprint recognition method and system based on local representation of fingerprints

Abstract

The application discloses a fingerprint recognition method and system based on local representation of fingerprints. The method first extracts fixed-length representation from the collected fingerprint image by using a convolutional neural network, then performs fingerprint comparison according to the extracted feature representation, and calculates the corresponding similarity score. In the application, the input of the fingerprint recognition system based on the fingerprint posture is various types of collected fingerprint images, and the output is the similarity score between fingerprints. The fingerprint recognition based on the fingerprint posture of the application can effectively cope with the interference of background information in the non-fingerprint area when extracting the fixed-length descriptor, and reduce the influence of the non-overlapping area on the recognition performance when comparing fingerprints of different types and different effective area sizes.

Description

Technical Field

[0001] This invention relates to the field of fingerprint recognition technology, and in particular to fingerprint recognition methods and systems based on local fingerprint representation. Background Technology

[0002] Thanks to the uniqueness and resistance to change of fingerprint features, fingerprints have been used in the field of identity recognition for thousands of years, making them one of the most widely used biometrics. They also have numerous applications in modern society, including access control verification, ID card registration, and immigration inspection. A fingerprint recognition system mainly consists of three modules: fingerprint image acquisition, feature extraction, and subsequent fingerprint comparison. Generally, fingerprints are represented as a set of key points, also known as minutiae. Due to the clear definition, random distribution, and relatively mature automatic extraction algorithms of minutiae, many existing fingerprint recognition systems still rely on minutiae for construction. However, the number of minutiae varies across different fingerprint images, representing a non-fixed-length feature representation. Therefore, fingerprint representation based on minutiae presents many problems, such as high computational cost of comparison algorithms, susceptibility to low-quality fingerprints, difficulty in template encryption, and the challenge of building end-to-end fingerprint recognition systems based on minutiae. Therefore, fixed-length fingerprint representations have received increasing attention. Because this type of fingerprint representation has a fixed feature dimension, similarity results between a large number of fingerprints can be quickly obtained using simple vector and matrix operations, effectively improving the accuracy and efficiency of fingerprint recognition systems.

[0003] Numerous algorithms exist for fixed-length fingerprint representation extraction and feature comparison. These methods generally suffer from the following limitations and shortcomings: Some algorithms extract global features from the fingerprint image, only effective when the fingerprint region is relatively complete and ideal. They fail to achieve good performance when the background noise is complex or the fingerprint region is incomplete. While some algorithms incorporate features such as fingerprint minutiae, they still ultimately use a global integration approach to obtain the global fingerprint representation, losing information such as the position and orientation of minutiae. This also makes them unsuitable for handling complex background noise or incomplete fingerprint regions. Empirical feature-based methods are highly sensitive to background noise and ridge quality, and their limited feature representation capabilities affect the algorithm's generalization ability. Summary of the Invention

[0004] The present invention aims to at least partially solve one of the technical problems in the related art.

[0005] To address this, the present invention proposes a fingerprint recognition method based on local fingerprint representation, which can extract a fixed-length local representation of fingerprint depth according to the fingerprint's posture, including the center position and orientation of the fingerprint, for comparison between various types of fingerprints, thereby improving the efficiency and accuracy of fingerprint recognition systems such as fingerprint retrieval and matching.

[0006] Another objective of this invention is to propose a fingerprint recognition system based on local fingerprint representation.

[0007] To achieve the above objectives, the present invention proposes a fingerprint recognition method based on local fingerprint representation, comprising:

[0008] Two-dimensional poses of the fingerprint image to be predicted and the fingerprint images in the fingerprint database are obtained respectively;

[0009] Based on the two-dimensional pose, fingerprint alignment is performed on the fingerprint image to be predicted and the fingerprint image in the fingerprint database, and the trained fingerprint local representation extraction network model is used to extract the fingerprint local representation of the aligned fingerprint image.

[0010] The effective dimension of the local fingerprint representation is constrained by the effective region to calculate the fingerprint similarity score between the fingerprint image to be predicted and fingerprint images in the fingerprint database.

[0011] Fingerprint recognition is performed on the fingerprint image to be predicted based on the fingerprint similarity score to obtain the fingerprint recognition result.

[0012] The fingerprint recognition method based on partial fingerprint representation according to embodiments of the present invention may also have the following additional technical features:

[0013] In one embodiment of the present invention, after extracting the local fingerprint representations of the aligned fingerprint images using a trained fingerprint extraction model, the method further includes:

[0014] The fingerprint local representation is quantized.

[0015] In one embodiment of the present invention, after calculating the fingerprint similarity score, the method further includes:

[0016] The fingerprint similarity score is standardized based on the size of the overlapping region between the fingerprint image to be predicted and the fingerprint images in the fingerprint database.

[0017] In one embodiment of the present invention, after calculating the fingerprint similarity score, the method further includes:

[0018] Before extracting the fingerprint local representations of the aligned fingerprint images using the trained fingerprint local representation extraction network model, the method further includes training the fingerprint local representation extraction network model, including:

[0019] Obtain aligned training fingerprint images; wherein, the training fingerprint images include sample images of fingerprint images to be predicted and sample images of fingerprint images in the fingerprint database;

[0020] The aligned training fingerprint image is input into a fingerprint local representation extraction network model based on a loss function for model training, so as to train the model parameters based on the loss calculation results of the sample images;

[0021] A trained fingerprint local representation extraction network model is obtained based on the trained model parameters.

[0022] In one embodiment of the present invention, obtaining the two-dimensional pose of the fingerprint image to be predicted includes:

[0023] A two-dimensional pose prediction result of the fingerprint image to be predicted is obtained by a fingerprint pose estimation algorithm; wherein, the two-dimensional pose prediction result includes the center position information and orientation information of the fingerprint.

[0024] To achieve the above objectives, another aspect of the present invention proposes a fingerprint recognition system based on local fingerprint representation, comprising:

[0025] The fingerprint 2D pose acquisition module is used to acquire the 2D pose of the fingerprint image to be predicted and the fingerprint image in the fingerprint database, respectively.

[0026] The fingerprint local representation extraction module is used to perform fingerprint alignment on the fingerprint image to be predicted and the fingerprint image in the fingerprint database based on the two-dimensional pose, and to extract the fingerprint local representation of the aligned fingerprint image using the trained fingerprint local representation extraction network model.

[0027] The fingerprint similarity score calculation module is used to constrain the effective dimension of the local fingerprint representation by the effective region to calculate the fingerprint similarity score between the fingerprint image to be predicted and the fingerprint images in the fingerprint database.

[0028] The fingerprint recognition result output module is used to perform fingerprint recognition on the fingerprint image to be predicted based on the fingerprint similarity score to obtain the fingerprint recognition result.

[0029] The fingerprint recognition method and system based on local fingerprint representation of this invention consists of two stages: local representation extraction based on fingerprint pose and fingerprint comparison based on local representation and effective region. In the first stage, the two-dimensional pose information of the fingerprint, including the center position and orientation, is obtained first. Then, a feature extraction module extracts the local representation of the fingerprint. In the second stage, based on the aforementioned local fingerprint representation, a comparison module calculates a similarity score between any two fingerprints. The obtained similarity score can be standardized or normalized based on the size of the overlapping region between the fingerprints. The first stage is the foundational step of this invention. By extracting the corresponding feature representation for each local region of the fingerprint and then combining them into a complete fixed-length representation, the original positional information can be preserved in the extracted fingerprint representation. It can also be combined with the effective region of the fingerprint to determine the effective dimension of the extracted feature representation. The second stage combines the extracted local representation and the effective region to constrain the feature dimensions for comparison during fingerprint comparison, thereby calculating a matching similarity score. The score can then be normalized based on the size of the overlapping region to alleviate the problem of confidence differences in similarity scores caused by different overlapping regions. This effectively addresses background information interference in non-fingerprint regions when extracting fixed-length descriptors and reduces the impact of non-overlapping regions on recognition performance when comparing fingerprints of different types and effective region sizes.

[0030] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0031] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0032] Figure 1 This is a flowchart of a fingerprint recognition method based on local fingerprint representation according to an embodiment of the present invention;

[0033] Figure 2 This is a flowchart of another fingerprint recognition method based on local fingerprint representation according to an embodiment of the present invention;

[0034] Figure 3 This is an example diagram of a two-dimensional fingerprint pose according to an embodiment of the present invention;

[0035] Figure 4 This is an example diagram of an aligned fingerprint image according to an embodiment of the present invention;

[0036] Figure 5 This is a flowchart of fingerprint partial representation extraction according to an embodiment of the present invention;

[0037] Figure 6 This is a diagram of a deep convolutional network structure for fingerprint local representation extraction according to an embodiment of the present invention;

[0038] Figure 7 This is a schematic diagram of a fingerprint local representation and its binarization result according to an embodiment of the present invention;

[0039] Figure 8 This is a structural diagram of a fingerprint recognition system based on local fingerprint representation according to an embodiment of the present invention. Detailed Implementation

[0040] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0041] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0042] The fingerprint recognition method and system based on local fingerprint representation according to embodiments of the present invention are described below with reference to the accompanying drawings.

[0043] First, the symbols that may appear in the embodiments of the present invention will be explained:

[0044] H, W: The two-dimensional length and width of the local representation;

[0045] d: Number of feature channels in the local representation;

[0046] f q f g , : Local representation of fingerprints, derived from the query fingerprint (q) and the database fingerprint (g), respectively;

[0047] b q b g The binarization results of the local fingerprint representation are derived from the query fingerprint (q) and the database fingerprint (g), respectively.

[0048] h q h g : The valid fingerprint area, which comes from the query fingerprint (q) and the database fingerprint (g);

[0049] m q m g The binarization results of the valid fingerprint region are derived from the query fingerprint (q) and the database fingerprint (g), respectively.

[0050] Logical XOR;

[0051] ∩: Logical AND;

[0052] <·>: Vector dot product;

[0053] ⊙: Vector / matrix dot product;

[0054] ‖·‖: Norm 2;

[0055] HD: Normalized Hamming distance;

[0056] s: Similarity score;

[0057] s o Area of ​​the overlapping region;

[0058] S o : Normalized parameter for the area of ​​the overlapping region;

[0059] Figure 1 This is a flowchart of a fingerprint recognition method based on local fingerprint representation according to an embodiment of the present invention.

[0060] like Figure 1 As shown, the method includes, but is not limited to, the following steps:

[0061] S1, obtain the two-dimensional pose of the fingerprint image to be predicted and the fingerprint image in the fingerprint database respectively;

[0062] S2, based on the two-dimensional pose, perform fingerprint alignment on the fingerprint image to be predicted and the fingerprint image in the fingerprint database respectively, and use the trained fingerprint local representation extraction network model to extract the fingerprint local representation of the aligned fingerprint image respectively;

[0063] S3, constrain the effective dimension of the fingerprint local representation by the effective region to calculate the fingerprint similarity score between the fingerprint image to be predicted and the fingerprint images in the fingerprint database;

[0064] S4, perform fingerprint recognition on the fingerprint image to be predicted based on the fingerprint similarity score to obtain the fingerprint recognition result.

[0065] It is understood that the fingerprint recognition based on fingerprint pose of this invention is divided into two modules: fingerprint local representation extraction based on fingerprint pose and fingerprint comparison based on local representation and effective region. The fingerprint local representation extraction stage based on fingerprint pose is divided into three steps: fingerprint alignment, fingerprint local representation extraction, and feature representation quantization. In the fingerprint alignment step, the two-dimensional pose of the fingerprint is first obtained, and then the fingerprint image is aligned based on this pose. In the fingerprint local representation extraction step, the local representation of the fingerprint is extracted using a local feature extraction module on the aligned fingerprint image. It is understood that in the feature representation quantization step (this step is optional and not mandatory), the fingerprint local representation extracted in the previous step can be processed to better meet the needs of subsequent storage and efficient computation. The fingerprint comparison based on fingerprint local representation and effective region is divided into two steps: similarity score calculation and similarity score standardization. In the similarity score calculation step, the effective dimension of the fingerprint local representation is constrained by the effective region, and the similarity score between fingerprints is calculated. In the similarity score standardization step (this step is optional), the similarity score is standardized based on the size of the overlapping region. Figure 2 As shown.

[0066] It is understood that the feature extraction model of the present invention can be an empirically designed model or a model trained from fingerprint data. The algorithm requires a large number of fingerprint images and corresponding fingerprint two-dimensional poses and finger identification labels. The parameters of the local representation extraction model are optimized using machine learning methods.

[0067] It is understood that this invention can be widely applied to fingerprint images of various types and effective areas, such as rolling fingerprints, planar fingerprints, on-site fingerprints, and non-contact fingerprints; it can also be applied to other biometric identification problems, such as palm prints, faces, irises, hand shapes, and palm veins. This invention is illustrated using fingerprint local representation extraction and fingerprint comparison as examples.

[0068] Specifically, the fingerprint pose-based local representation extraction in this invention consists of three steps: fingerprint alignment, fingerprint local representation extraction, and feature representation quantization. The fingerprint image is aligned according to its two-dimensional pose. Local representations are then extracted from the aligned fingerprint image. This incorporates prior knowledge of fingerprint feature distribution, effectively reducing the complexity of model learning, facilitating subsequent calculation of overlapping fingerprint regions, and allowing for numerical quantization of the extracted fingerprint feature representations, thus reducing the storage and computational resource requirements.

[0069] In one example, there are many definitions of the two-dimensional fingerprint pose used in the fingerprint alignment step of this invention. Here, only one of them is used as an example, such as... Figure 3As shown. There are many methods for obtaining the two-dimensional pose from a fingerprint image, such as manual specification or automatic estimation using algorithms. Here, we take automatic estimation using algorithms as an example. A fingerprint pose estimation algorithm obtains the two-dimensional pose prediction result of any fingerprint, including the center position and orientation of the fingerprint. Then, the fingerprint image is aligned based on the center position and orientation information. Figure 4 An example of the result after fingerprint pair alignment is shown.

[0070] In one example, the fingerprint local representation extraction step in this invention can use many approaches, including empirically designed models and machine learning-based models. Here, we take a deep convolutional neural network using supervised learning as an example.

[0071] Figure 5 The flowchart for fingerprint local representation extraction is shown. During offline processing, a deep convolutional network is trained using a large number of aligned fingerprint images and corresponding finger identification labels to obtain optimized network parameters.

[0072] For example, an aligned training fingerprint image is obtained; wherein, the training fingerprint image includes a sample image of the fingerprint image to be predicted and a sample image of the fingerprint image in the fingerprint database;

[0073] The aligned training fingerprint images are input into a fingerprint local representation extraction network model based on a loss function for model training, so as to train the model parameters based on the loss calculation results of the sample images;

[0074] A trained fingerprint local representation extraction network model is obtained based on the trained model parameters.

[0075] Specifically, a feature extraction module based on a deep convolutional network extracts a feature representation of size W×H from the fingerprint image. Each location is a d-dimensional feature vector, corresponding to a local region in the original fingerprint image. Taking H=W=16 and d=6 as an example, the final fingerprint local representation has a dimension of 16×16×6. Furthermore, this module also extracts the corresponding effective fingerprint region, which has the same two-dimensional resolution as the aforementioned local representation. Figure 6 An example of a deep convolutional network structure for fingerprint local representation extraction is shown.

[0076] It is understood that the feature representation quantization step in this invention can use many schemes, such as quantizing it into an unsigned 8-bit integer. Here, we take the binarization of the feature representation as an example. For the local representation extracted in the previous step, the feature representation is binarized into 0 and 1 representations according to whether its value is greater than 0, where 0 indicates that the original value is less than 0, and 1 indicates that the original value is greater than or equal to 0. Figure 7 Examples of the original local representation extracted in the previous step and the result after binarization are shown respectively.

[0077] Furthermore, the fingerprint matching based on local representation and effective regions in this invention consists of two steps: similarity calculation and similarity score standardization. Based on the local fingerprint representation obtained in the previous stage, the similarity score between any two fingerprints can be quickly obtained through simple vector and matrix operations. Moreover, based on the different sizes of overlapping regions, the similarity score can be standardized to mitigate the impact of low-confidence similarity scores on the final matching performance.

[0078] The similarity calculation steps in this invention are related to the feature representations used. For ease of description and understanding, the original fingerprint local representation and the binarized result are used as examples.

[0079] There are many schemes for similarity calculation of the original fingerprint local representation extracted in the previous stage. Here, we take cosine similarity as an example. The specific calculation formula is as follows:

[0080]

[0081] Where f′ q and f′ g The feature representation after fingerprint effective region constraint is calculated using the following formulas:

[0082] f′ q =f q ⊙h q+g f′ g =f g ⊙h q+g

[0083] And h q+g The overlapping area between fingerprints is represented by the following formula:

[0084] h q+g =h q ⊙h g

[0085] For the local representation after binarization, there are still many options for similarity calculation. For example, the cosine similarity mentioned above can still be used, but here we take the simpler Hamming distance as an example. The specific calculation formula is as follows:

[0086]

[0087] The purpose of the similarity score standardization step in this invention is to mitigate the differences in similarity score confidence caused by varying overlap region sizes, thereby reducing its impact on matching performance. The similarity score standardization scheme varies depending on the feature representation used in the previous step. This is because the appropriate standardization method needs to be selected based on the score distribution model obtained from different feature representations. Here, we take the use of binarized feature representations as an example. At this point, the false match score distribution between fingerprints is a binomial distribution. By normalizing the variance, the formula for similarity score standardization is:

[0088]

[0089] Note that the similarity score standardization step is not mandatory. When the variation in the overlapping area is not significant, the calculated confidence scores will not differ considerably, so the standardization step can be omitted.

[0090] The fingerprint recognition method based on fingerprint local representation according to embodiments of the present invention can extract a fixed-length fingerprint depth local representation according to the fingerprint posture, including the center position and orientation of the fingerprint, for comparison between various types of fingerprints, thereby improving the efficiency and accuracy of fingerprint retrieval, matching and other fingerprint recognition systems.

[0091] To achieve the above embodiments, such as Figure 8 As shown, this embodiment also provides a fingerprint recognition system 10 based on fingerprint local representation. The system 10 includes a fingerprint two-dimensional pose acquisition module 100, a fingerprint local representation extraction module 200, a fingerprint similarity score calculation module 300, and a fingerprint recognition result output module 400.

[0092] The fingerprint two-dimensional pose acquisition module 100 is used to acquire the two-dimensional poses of the fingerprint image to be predicted and the fingerprint image in the fingerprint database, respectively.

[0093] The fingerprint local representation extraction module 200 is used to perform fingerprint alignment between the fingerprint image to be predicted and the fingerprint image in the fingerprint database based on two-dimensional pose, and to extract the fingerprint local representation of the aligned fingerprint image using a trained fingerprint local representation extraction network model.

[0094] The fingerprint similarity score calculation module 300 is used to constrain the effective dimension of the local representation of the fingerprint by the effective region to calculate the fingerprint similarity score between the fingerprint image to be predicted and the fingerprint images in the fingerprint database.

[0095] The fingerprint recognition result output module 400 is used to perform fingerprint recognition on the fingerprint image to be predicted based on the fingerprint similarity score to obtain the fingerprint recognition result.

[0096] Furthermore, after the fingerprint local representation extraction module 200, the system 10 also includes a fingerprint local representation quantization module, used for:

[0097] The local representation of the fingerprint is quantized.

[0098] Furthermore, following the fingerprint similarity score calculation module 300, the system 10 also includes a standardization processing module for:

[0099] The fingerprint similarity score is standardized based on the size of the overlapping region between the fingerprint image to be predicted and the fingerprint images in the fingerprint database.

[0100] Furthermore, prior to the fingerprint local representation extraction module 200, the system 10 also includes a network model training module for:

[0101] Obtain aligned training fingerprint images; wherein, training fingerprint images include sample images of fingerprint images to be predicted and sample images of fingerprint images in the fingerprint database;

[0102] The aligned training fingerprint images are input into a fingerprint local representation extraction network model based on a loss function for model training, so as to train the model parameters based on the loss calculation results of the sample images;

[0103] A trained fingerprint local representation extraction network model is obtained based on the trained model parameters.

[0104] Furthermore, the fingerprint two-dimensional pose acquisition module 100 is also used for:

[0105] The fingerprint pose estimation algorithm is used to obtain the two-dimensional pose prediction result of the fingerprint image to be predicted; the two-dimensional pose prediction result includes the center position information and orientation information of the fingerprint.

[0106] According to the fingerprint recognition system based on fingerprint local representation according to the embodiments of the present invention, a fingerprint depth local representation of a fixed length can be extracted according to the fingerprint posture, including the center position and orientation of the fingerprint, for comparison between various types of fingerprints, thereby improving the efficiency and accuracy of fingerprint recognition systems such as fingerprint retrieval and matching.

[0107] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0108] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A fingerprint recognition method based on local fingerprint representation, characterized in that, The method includes the following steps: Two-dimensional poses of the fingerprint image to be predicted and the fingerprint images in the fingerprint database are obtained respectively; Based on the two-dimensional pose, fingerprint alignment is performed on the fingerprint image to be predicted and the fingerprint image in the fingerprint database, and the fingerprint local representation extraction network model is trained to extract the fingerprint local representation of the aligned fingerprint image, and the fingerprint local representation is quantized. The effective dimension of the local fingerprint representation is constrained by the effective region to calculate the fingerprint similarity score between the fingerprint image to be predicted and fingerprint images in the fingerprint database. Fingerprint recognition is performed on the fingerprint image to be predicted based on the fingerprint similarity score to obtain the fingerprint recognition result; The method further includes: The fingerprint pose estimation algorithm is used to obtain the two-dimensional pose of the fingerprint, including the center position and orientation, and the fingerprint image is aligned based on the pose. The local representation of the aligned fingerprint image is extracted, and the prior distribution knowledge of fingerprint features is introduced to reduce the learning complexity of the model. In offline mode, a deep convolutional network is trained using a large number of aligned fingerprint images and corresponding finger identification data to obtain optimized network parameters. Feature representation quantization involves quantizing the extracted local representations and performing binarization based on the relationship between the numerical value and 0: values ​​less than 0 are marked as 0, and values ​​greater than or equal to 0 are marked as 1. Fingerprint comparison is performed by combining the local representation of the fingerprint with the effective region, and similarity calculation and similarity score standardization are carried out in sequence. The similarity score between fingerprints is quickly solved by vector and matrix operations, and the similarity score standardization is completed by combining the size of the fingerprint overlap region.

2. The method according to claim 1, characterized in that, After calculating the fingerprint similarity score, the method further includes: The fingerprint similarity score is standardized based on the size of the overlapping region between the fingerprint image to be predicted and the fingerprint images in the fingerprint database.

3. The method according to claim 1, characterized in that, Before extracting the fingerprint local representations of the aligned fingerprint images using the trained fingerprint local representation extraction network model, the method further includes training the fingerprint local representation extraction network model, including: Obtain aligned training fingerprint images; wherein, the training fingerprint images include sample images of fingerprint images to be predicted and sample images of fingerprint images in the fingerprint database; The aligned training fingerprint image is input into a fingerprint local representation extraction network model based on a loss function for model training, so as to train the model parameters based on the loss calculation results of the sample images; A trained fingerprint local representation extraction network model is obtained based on the trained model parameters.

4. The method according to claim 1, characterized in that, Obtain the two-dimensional pose of the fingerprint image to be predicted, including: A two-dimensional pose prediction result of the fingerprint image to be predicted is obtained by a fingerprint pose estimation algorithm; wherein, the two-dimensional pose prediction result includes the center position information and orientation information of the fingerprint.

5. A fingerprint recognition system based on local fingerprint representation, characterized in that, include: The fingerprint 2D pose acquisition module is used to acquire the 2D pose of the fingerprint image to be predicted and the fingerprint image in the fingerprint database, respectively. The fingerprint local representation extraction module is used to perform fingerprint alignment on the fingerprint image to be predicted and the fingerprint image in the fingerprint database based on the two-dimensional pose, and to extract the fingerprint local representation of the aligned fingerprint image using the trained fingerprint local representation extraction network model. A fingerprint local representation quantization module is used to: perform quantization processing on the fingerprint local representation; The fingerprint similarity score calculation module is used to constrain the effective dimension of the local fingerprint representation by the effective region to calculate the fingerprint similarity score between the fingerprint image to be predicted and the fingerprint images in the fingerprint database. A fingerprint recognition result output module is used to perform fingerprint recognition on the fingerprint image to be predicted based on the fingerprint similarity score to obtain a fingerprint recognition result. The system is also used for: The fingerprint pose estimation algorithm is used to obtain the two-dimensional pose of the fingerprint, including the center position and orientation, and the fingerprint image is aligned based on the pose. The local representation of the aligned fingerprint image is extracted, and the prior distribution knowledge of fingerprint features is introduced to reduce the learning complexity of the model. In offline mode, a deep convolutional network is trained using a large number of aligned fingerprint images and corresponding finger identification data to obtain optimized network parameters. Feature representation quantization involves quantizing the extracted local representations and performing binarization based on the relationship between the numerical value and 0: values ​​less than 0 are marked as 0, and values ​​greater than or equal to 0 are marked as 1. Fingerprint comparison is performed by combining the local representation of the fingerprint with the effective region, and similarity calculation and similarity score standardization are carried out in sequence. The similarity score between fingerprints is quickly solved by vector and matrix operations, and the similarity score standardization is completed by combining the size of the fingerprint overlap region.

6. The system according to claim 5, characterized in that, Following the fingerprint similarity score calculation module, the system further includes a standardization processing module, used for: The fingerprint similarity score is standardized based on the size of the overlapping region between the fingerprint image to be predicted and the fingerprint images in the fingerprint database.

7. The system according to claim 5, characterized in that, Prior to the fingerprint local representation extraction module, the system further includes a network model training module, used for: Obtain aligned training fingerprint images; wherein, the training fingerprint images include sample images of fingerprint images to be predicted and sample images of fingerprint images in the fingerprint database; The aligned training fingerprint image is input into a fingerprint local representation extraction network model based on a loss function for model training, so as to train the model parameters based on the loss calculation results of the sample images; A trained fingerprint local representation extraction network model is obtained based on the trained model parameters.

8. The system according to claim 5, characterized in that, The fingerprint two-dimensional pose acquisition module is also used for: A two-dimensional pose prediction result of the fingerprint image to be predicted is obtained by a fingerprint pose estimation algorithm; wherein, the two-dimensional pose prediction result includes the center position information and orientation information of the fingerprint.

Images