A model training method, a fingerprint image processing method, a device and an electronic device

Abstract

The application provides a model training method, a fingerprint image processing method, a device and electronic equipment. In the model training method, pre-enhanced and post-enhanced non-contact fingerprint images are input into a model, a feature alignment constraint is introduced in a convolution network part of the model, a geometric alignment constraint is introduced in an STN layer, image features of the non-contact fingerprint and a contact fingerprint are respectively extracted by using a pre-trained model, and the two features are constrained to be similar. Based on the three-part constraint solution, a fingerprint image processing model is trained to obtain an optimal model parameter solution, the model convergence is effectively accelerated, and the generalization ability of the model is improved.

Description

Technical Field

[0001] This invention belongs to the field of image processing, specifically relating to a model training method, a fingerprint image processing method, a device, and an electronic device. Background Technology

[0002] Fingerprint recognition is a crucial application in the field of biometrics. Past applications primarily relied on contact fingerprinting, but this method has significant drawbacks, such as hygiene threats, sensor degradation due to continuous physical contact, and potential fingerprint threats. The widespread adoption of contact fingerprinting has resulted in a large accumulation of fingerprints across various scenarios, including access control, attendance systems, and identity authentication. Re-collecting contactless fingerprints would be extremely resource-intensive, necessitating contactless fingerprint products that are compatible with existing contact fingerprint systems.

[0003] The main differences between contact and contactless fingerprinting lie in two aspects: first, contactless fingerprinting introduces perspective distortion during the acquisition process; second, contact fingerprinting causes image distortion due to pressure. Therefore, various methods have been proposed to correct these two distortions. One correction method involves extracting features from contactless fingerprints to generate a corresponding transformation matrix and using a Spatial Transformation Network (STN) for spatial transformation. However, since there are no perfectly registered contact and contact fingerprint image pairs, training image processing models for contactless fingerprints relies on feature extraction from contact fingerprints. Therefore, it cannot effectively guide the learning of image information during the training process of network models that use contactless fingerprint images as input. Summary of the Invention

[0004] Based on this, the present invention proposes a model training method, a fingerprint image processing method, a device, and an electronic device, which utilizes enhanced image constraints on model parameters to achieve a higher matching degree between non-contact fingerprint images and contact fingerprints after model processing.

[0005] In a first aspect, the present invention provides a model training method, wherein the fingerprint image processing model trained by the method is used to process contactless fingerprint images, comprising:

[0006] Obtain the raw image of the contactless fingerprint;

[0007] Perform data augmentation transformation on the original image to obtain the enhanced image and the geometric transformation matrix;

[0008] The original image and the enhanced image are respectively input into the convolutional network of the fingerprint image processing model. Feature alignment constraints are applied to the output based on the original image and the output based on the enhanced image to obtain the first constraint solution.

[0009] The output based on the original image and the output based on the enhanced image are respectively input into the STN layer of the fingerprint image processing model to obtain the first transformed image and the second transformed image. Geometric alignment constraints are applied to the first transformed image and the second transformed image to obtain the second constraint solution.

[0010] The pre-trained model is used to extract features from the first transformed image and the original image of the contact fingerprint, and feature constraints are applied to the non-contact fingerprint features and the contact fingerprint features to obtain the third constraint solution;

[0011] Based on the first, second, and third constraint solutions, the parameters of the fingerprint image processing model are optimized to obtain a trained fingerprint image processing model.

[0012] Preferably, the original image and the enhanced image are respectively input into the convolutional network of the fingerprint image processing model, and feature alignment constraints are applied to the output based on the original image and the output based on the enhanced image, including:

[0013] A convolutional network consists of three cascaded layers (CNN).

[0014] The original image and the enhanced image are input into three layers of CNN sequentially. The first CNN layer produces the first feature map of the original image and the second feature map of the enhanced image.

[0015] The first feature map is subjected to data augmentation transformation based on the geometric transformation matrix to obtain the third feature map;

[0016] By applying feature alignment constraints to the second and third feature maps, the first constraint solution is obtained.

[0017] Preferably, the above method further includes:

[0018] Apply the first constraint solution to the second and / or third CNN layers.

[0019] Preferably, the output based on the original image and the output based on the enhanced image are respectively input into the STN layer of the fingerprint image processing model to obtain a first transformed image and a second transformed image, including:

[0020] The output based on the original image and the output based on the enhanced image are processed through an STN layer to obtain a first texture map and a second texture map, respectively. The first texture map and the second texture map are then applied to the original image and the enhanced image, respectively, to obtain a first transformed image and a second transformed image.

[0021] Preferably, a data augmentation layer is cascaded after the STN layer of the fingerprint image processing model, and the input of the first transformed image into the pre-trained model includes:

[0022] The first transformed image is input into the data augmentation layer for image augmentation, and the augmented first transformed image is input into the pre-trained model.

[0023] Preferably, based on the first constraint solution, the second constraint solution, and the third constraint solution, the parameters of the fingerprint image processing model are optimized, including:

[0024] Weights are assigned to the first, second, and third constraint solutions respectively, so that the constraint expression of the fingerprint image processing model is as follows:

[0025]

[0026] This represents the overall constraint solution for the fingerprint image processing model. , and These represent the solutions to the first, second, and third constraints, respectively. , and The weights correspond to the first constraint solution, the second constraint solution, and the third constraint solution, respectively.

[0027] Based on the above constraint expression calculation, the Adam optimizer is used to optimize the model parameters to obtain a trained fingerprint image processing model.

[0028] In a second aspect, the present invention provides a fingerprint image processing method, comprising:

[0029] Obtain the raw image of the contactless fingerprint;

[0030] The fingerprint image processing model trained using the model training method in the first aspect processes the original image of the non-contact fingerprint to obtain the processed non-contact fingerprint image, which is then used for matching with the contact fingerprint image.

[0031] Thirdly, the present invention provides a model training apparatus, comprising:

[0032] Image acquisition unit, used to acquire the original image of the contactless fingerprint;

[0033] The image enhancement unit is used to perform data enhancement transformations on the original image to obtain the enhanced image and the geometric transformation matrix;

[0034] The image processing unit, which includes a pre-trained model and a fingerprint image processing model to be trained, is used for:

[0035] The original image and the enhanced image are respectively input into the convolutional network of the fingerprint image processing model. Feature alignment constraints are applied to the output based on the original image and the output based on the enhanced image to obtain the first constraint solution.

[0036] The output based on the original image and the output based on the enhanced image are respectively input into the STN network of the fingerprint image processing model to obtain the first transformed image and the second transformed image. Geometric alignment constraints are applied to the first transformed image and the second transformed image to obtain the second constraint solution.

[0037] The pre-trained model is used to extract features from the first transformed image and the original image of the contact fingerprint, and feature constraints are applied to the non-contact fingerprint features and the contact fingerprint features to obtain the third constraint solution;

[0038] Based on the first, second, and third constraint solutions, the parameters of the fingerprint image processing model are optimized to obtain a trained fingerprint image processing model.

[0039] Fourthly, the present invention provides a fingerprint image processing apparatus, comprising:

[0040] Image acquisition unit, used to acquire the original image of the contactless fingerprint;

[0041] The image processing unit is used to process the original image of the non-contact fingerprint using a fingerprint image processing model trained by the model training device of the third aspect, to obtain a processed non-contact fingerprint image for matching with a contact fingerprint image.

[0042] Fifthly, the present invention provides an electronic device, comprising: a memory and a processor;

[0043] The memory is used to store programs;

[0044] The processor is configured to invoke a program stored in the memory to execute the model training method provided in the first aspect embodiment and / or any possible implementation in combination with the first aspect embodiment, or to execute the fingerprint image processing method provided in the second aspect embodiment.

[0045] In a sixth aspect, the present invention also provides a readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the model training method provided by the first aspect embodiment and / or any possible implementation in conjunction with the first aspect embodiment, or executes the fingerprint image processing method provided by the second aspect embodiment.

[0046] The beneficial effects of this invention are as follows:

[0047] The fingerprint image processing model obtained by the model training method proposed in this invention introduces feature alignment constraints in the convolutional network part of the model, constraining the similarity of features between non-contact fingerprint images before and after constraint enhancement. Geometric alignment constraints are introduced in the spatial transformation network (STN) part of the model, constraining the similarity between non-contact fingerprint images before and after constraint enhancement through STN transformation. Furthermore, the image features of non-contact fingerprints and contact fingerprints are extracted using a pre-trained model, constraining the similarity between the two. The three constraint solutions are used to optimize the model parameters of the fingerprint image processing model, so that the non-contact fingerprint images processed by the model can be directly used for matching with contact fingerprints, effectively accelerating model convergence, improving the model's generalization ability, and simultaneously improving the matching accuracy between non-contact fingerprints and contact fingerprints. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of a model training method provided in one embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of a model training method provided in another embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of a model training method that introduces feature alignment constraints in the first layer of a CNN according to an embodiment of the present invention.

[0052] Figure 4 This is a schematic diagram of a model training method that introduces feature alignment constraints in the first and second CNN layers of the model according to an embodiment of the present invention.

[0053] Figure 5 This is a schematic diagram of a model training method that introduces feature alignment constraints in the first and third CNN layers of the model according to an embodiment of the present invention.

[0054] Figure 6 This is a schematic diagram of a model training method that introduces feature alignment constraints in all three layers of a CNN model, according to an embodiment of the present invention.

[0055] Figure 7 This is a schematic diagram of a model training method provided in another embodiment of the present invention;

[0056] Figure 8 It is to utilize Figure 7A schematic diagram illustrating the non-contact fingerprint image processing of the model trained using the illustrative model training method.

[0057] Figure 9 This is a structural diagram of the model training device provided in an embodiment of the present invention;

[0058] Figure 10 This is a structural diagram of the fingerprint image processing device provided in an embodiment of the present invention;

[0059] Figure 11 This is an electronic device architecture diagram provided in an embodiment of the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0061] See Figure 1 This embodiment provides a model training method for a fingerprint image processing model, including:

[0062] Obtain the raw image of the contactless fingerprint.

[0063] Specifically, the raw non-contact fingerprint images obtained here are used to train the model. The raw images often have deformation defects, and if they are directly used to match contact fingerprints, the accuracy will be low and the efficiency will be low.

[0064] The original image is subjected to data augmentation transformation to obtain the enhanced image and the geometric transformation matrix.

[0065] In an optional implementation, the enhancement transformation of the original image can be a geometric deformation, such as affine transformation, perspective transformation, mirroring, rotation, scaling, noise increase, etc. The geometric transformation matrix records the specific transformation information, such as coordinates, vectors, etc.

[0066] The original image and the enhanced image are respectively input into the convolutional network of the fingerprint image processing model. Feature alignment constraints are applied to the output based on the original image and the output based on the enhanced image to obtain the first constraint solution.

[0067] The output based on the original image and the output based on the enhanced image are respectively input into the STN layer of the fingerprint image processing model to obtain the first transformed image and the second transformed image. Geometric alignment constraints are applied to the first transformed image and the second transformed image to obtain the second constraint solution.

[0068] Specifically, this can be understood as using the same model to process non-contact fingerprint images before and after enhancement. The trained image processing model includes a convolutional network and a spatial transformation network (STN). The convolutional network introduces feature alignment constraints to ensure the feature similarity of the non-contact fingerprint images before and after enhancement. The STN introduces geometric alignment constraints. The role of the STN is to generate a texture map, which records the parameters for geometric transformations such as translation, scaling, and rotation of the original image. The texture map is applied to the input image to obtain the transformed image. Therefore, this part mainly constrains the geometric similarity of the transformed images obtained by the STN from the non-contact fingerprint images before and after enhancement.

[0069] The pre-trained model is used to extract features from the first transformed image and the original image of the contact fingerprint. Feature constraints are applied to the non-contact fingerprint features and the contact fingerprint features to obtain the third constraint solution.

[0070] Specifically, the pre-trained model here is a feature extraction model trained on a large number of contact fingerprint images. The pre-trained model is used to extract features from non-contact fingerprints and contact fingerprints respectively, and the features of the two are constrained to be similar. The non-contact fingerprint image input to the pre-trained model is the image transformed by the fingerprint image processing model mentioned above. This constraint solution is used to optimize the parameters of the fingerprint image processing model.

[0071] In an optional implementation, the fingerprint image processing model further includes an enhancement layer concatenated after the STN layer. The non-contact fingerprint image, after being transformed by the convolutional network and the STN layer, is enhanced by the enhancement layer before entering the pre-trained model. The training process of the model with the enhancement layer is as follows: Figure 2 As shown.

[0072] Based on the first, second, and third constraint solutions, the parameters of the fingerprint image processing model are optimized to obtain a trained fingerprint image processing model.

[0073] Specifically, each of the three constraint solutions mentioned above has a corresponding weight, so the total constraint solution of the model can be calculated. The total constraint solution is applied to the parameter optimization process of the model. Specifically, the commonly used Adam optimizer can be used to solve the model parameters until the optimal model parameters are obtained.

[0074] In an optional implementation, the constraint solution can be obtained using L1 constraints, L2 constraints, or cosine constraints.

[0075] See Figure 3 This embodiment provides another model training method, which is similar to... Figure 1The difference between the illustrated embodiment and the traditional one is that the convolutional network in this embodiment consists of three CNN layers connected in series. The contactless fingerprint images before and after enhancement are processed by the first CNN layer to obtain a first feature map and a second feature map, respectively. Constraining the first and second feature maps is similar to obtaining a first constrained solution. The remaining parts are the same as... Figure 1 The illustrated embodiments are similar and can be referred to in the foregoing descriptions, which will not be repeated here.

[0076] In an optional implementation, for the first feature map obtained by the original image entering the first layer of the CNN, the same enhancement transformation is performed on the first feature map according to the geometric transformation matrix for data augmentation of the original image to obtain the transformed third feature map. The third feature map is constrained to be similar to the second feature map, but the similarity between the first feature map and the second feature map is not directly constrained.

[0077] In optional implementations, feature alignment constraints can also be applied to the second or third CNN layer, or simultaneously to all three CNN layers, such as... Figures 4-6 As shown.

[0078] Understandable, Figures 3-6 The model training process shown can also include Figure 2 The enhancement layers shown are not listed exhaustively here.

[0079] Based on the above embodiments, the present invention will be further described below through a specific training process.

[0080] like Figure 7 As shown, the fingerprint image processing model in this embodiment consists of three layers: a CNN layer, an STN layer, and an enhancement layer. The model parameters are shared when images before and after enhancement are input into the model. The model training process is as follows:

[0081] S101. Perform data augmentation transformation on the original contactless fingerprint image img_A0. For example, perspective transformation, scaling, and noise addition are performed to obtain the enhanced image img_A0_aug, and the corresponding geometric transformation matrix points_map. points_map records the specific transformations performed on img_A0. .

[0082] S102. Input img_A0 and img_A0_aug into three CNN layers respectively. After passing through the first CNN layer, the first feature map ref_feature_layer1 of the original image and the second feature map aug_feature_layer1 of the enhanced image are obtained respectively. According to the points_map, the same enhancement transformation is performed on the first feature map ref_feature_layer1. The transformed third feature map ref_feature_layer1_aug is obtained. The constraints of ref_feature_layer1_aug are similar to those of aug_feature_layer1. Here, cosine constraints are used, and the constraint solution is expressed as follows:

[0083]

[0084] S103. Apply the constraint solution obtained in step S102 to the third CNN layer.

[0085] S104. The feature maps obtained after the original image img_A0 and the enhanced image img_A0_aug pass through three CNN layers are then fed into STN layers. The role of the STN layer is to generate a texture map, which is used to transform the image input to the STN layer. Specifically, based on the input of the original and enhanced images, a first texture map ref_grid_map and a second texture map aug_grid_map are generated respectively, and applied to img_A0 and img_A0_aug respectively, resulting in a first transformed image img_A0_grid and a second transformed image img_A0_aug_grid. The constraints on both are similar; the geometric alignment constraint here uses L1 constraint, and the constraint solution is expressed as follows:

[0086]

[0087] S105. Input the first transformed image into the enhancement layer for image enhancement to obtain the enhanced transformed image img_A0_final.

[0088] S106. Input the transformed image img_A0_final and the original contact fingerprint image img_B0 into the pre-trained model for feature extraction. The pre-trained model is a feature extraction model trained on a large number of contact fingerprint images, which obtains the non-contact fingerprint feature img_A0_final_feature and the contact fingerprint feature img_B0_feature, respectively. Apply cosine constraints to the two features, and the constraint solution is expressed as follows:

[0089]

[0090] S107. Based on the constraint solutions obtained in steps S102, S104, and S106, the overall constraint expression of the fingerprint image processing model is as follows:

[0091]

[0092] , and The weights of each constraint solution are represented by the Adam optimizer, which is used to solve for the model parameters until the optimal model parameters are obtained, thus completing the model training.

[0093] The above model training method can yield a fingerprint image processing model with better generalization ability. Figure 8 The process of non-contact fingerprint image processing using a trained model is shown.

[0094] By inputting any contactless fingerprint image into the model, a converted image can be obtained. The converted image can be directly applied to any contact fingerprint algorithm for contact fingerprint matching.

[0095] This application embodiment also provides a model training device 100, such as Figure 9 As shown, it includes:

[0096] Image acquisition unit 110 is used to acquire the original image of the contactless fingerprint.

[0097] The image enhancement unit 120 is used to perform data enhancement transformation on the original image to obtain an enhanced image and a geometric transformation matrix.

[0098] The image processing unit 130 has a pre-trained model and a fingerprint image processing model to be trained.

[0099] The model training device 100 provided in this application embodiment has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0100] This application also provides a fingerprint image processing device 200, such as... Figure 10 As shown, it includes:

[0101] Image acquisition unit 210 is used to acquire the original image of the contactless fingerprint;

[0102] The image processing unit 220 is used to process the original image of the non-contact fingerprint using the fingerprint image processing model trained by the model training device 100 described above, and obtain the processed non-contact fingerprint image for matching with the contact fingerprint image.

[0103] The fingerprint image processing device 200 provided in this application embodiment has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0104] See Figure 11It shows a hardware structure block diagram of an electronic device, including: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;

[0105] In this embodiment of the application, the number of processor 1, communication interface 2, memory 3, and communication bus 4 is at least one, and processor 1, communication interface 2, and memory 3 communicate with each other through communication bus 4;

[0106] Processor 1 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0107] Memory 3 may include high-speed RAM, and may also include non-volatile memory, such as at least one disk storage device;

[0108] The memory stores a program, which the processor can call. The program is used to: implement the various processing steps of the aforementioned model training method, or implement the various processing steps of the aforementioned disk failure prediction method.

[0109] This application also provides a readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the model training method provided in the above embodiments, or executes the fingerprint image processing method provided in the above embodiments.

[0110] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0111] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A model training method, characterized in that, The fingerprint image processing model trained by the method is used to process contactless fingerprint images, including: Obtain the raw image of the contactless fingerprint; The original image is subjected to data augmentation transformation to obtain an enhanced image and a geometric transformation matrix; The original image and the enhanced image are respectively input into the convolutional network of the fingerprint image processing model. Feature alignment constraints are applied to the output based on the original image and the output based on the enhanced image to obtain the first constraint solution. The output based on the original image and the output based on the enhanced image are respectively input into the STN layer of the fingerprint image processing model to obtain a first transformed image and a second transformed image. Geometric alignment constraints are applied to the first transformed image and the second transformed image to obtain a second constraint solution. The pre-trained model is used to extract features from the first transformed image and the original image of the contact fingerprint, and feature constraints are applied to the non-contact fingerprint features and the contact fingerprint features to obtain the third constraint solution; Based on the first constraint solution, the second constraint solution, and the third constraint solution, the parameters of the fingerprint image processing model are optimized to obtain a trained fingerprint image processing model.

2. The model training method according to claim 1, characterized in that, The original image and the enhanced image are respectively input into the convolutional network of the fingerprint image processing model. Feature alignment constraints are applied to the output based on the original image and the output based on the enhanced image to obtain the first constraint solution, including: The convolutional network comprises three cascaded CNN layers; The original image and the enhanced image are sequentially input into a three-layer CNN. The first feature map of the original image and the second feature map of the enhanced image are obtained in the first CNN layer. The first feature map is subjected to data augmentation transformation based on the geometric transformation matrix to obtain the third feature map; The second and third feature maps are subjected to feature alignment constraints to obtain the first constraint solution.

3. The model training method according to claim 2, characterized in that, The method further includes: Apply the first constraint solution to the second and / or third CNN layers.

4. The model training method according to claim 1, characterized in that, The output based on the original image and the output based on the enhanced image are respectively input into the STN layer of the fingerprint image processing model to obtain a first transformed image and a second transformed image, including: The output based on the original image and the output based on the enhanced image are processed through an STN layer to obtain a first texture map and a second texture map, respectively. The first texture map and the second texture map are then applied to the original image and the enhanced image, respectively, to obtain the first transformed image and the second transformed image.

5. The model training method according to claim 1, characterized in that, The fingerprint image processing model includes a data augmentation layer connected in series after the STN layer, and before inputting the first transformed image into the pre-trained model, it includes: The first transformed image is input into the data augmentation layer for image enhancement, and the enhanced first transformed image is used as the input to the pre-trained model.

6. The model training method according to claim 1, characterized in that, Based on the first constraint solution, the second constraint solution, and the third constraint solution, the parameters of the fingerprint image processing model are optimized, including: Weights are assigned to the first, second, and third constraint solutions respectively, so that the constraint expression of the fingerprint image processing model is as follows: ； This represents the overall constraint solution for the fingerprint image processing model. , and These represent the solutions to the first, second, and third constraints, respectively. , and The weights correspond to the first constraint solution, the second constraint solution, and the third constraint solution, respectively. Based on the constraints, the Adam optimizer is used to optimize the model parameters to obtain a trained fingerprint image processing model.

7. A fingerprint image processing method, characterized in that, include: Obtain the raw image of the contactless fingerprint; The fingerprint image processing model trained using the model training method described in any one of claims 1 to 6 is used to process the original image of the non-contact fingerprint to obtain a processed non-contact fingerprint image, which is then used for matching with a contact fingerprint image.

8. A model training device, characterized in that, include: Image acquisition unit, used to acquire the original image of the contactless fingerprint; The image enhancement unit is used to perform data enhancement transformations on the original image to obtain the enhanced image and the geometric transformation matrix; The image processing unit, which includes a pre-trained model and a fingerprint image processing model to be trained, is used for: The original image and the enhanced image are respectively input into the convolutional network of the fingerprint image processing model. Feature alignment constraints are applied to the output based on the original image and the output based on the enhanced image to obtain the first constraint solution. The output based on the original image and the output based on the enhanced image are respectively input into the STN network of the fingerprint image processing model to obtain the first transformed image and the second transformed image. Geometric alignment constraints are applied to the first transformed image and the second transformed image to obtain the second constraint solution. The pre-trained model is used to extract features from the first transformed image and the original image of the contact fingerprint, and feature constraints are applied to the non-contact fingerprint features and the contact fingerprint features to obtain the third constraint solution; Based on the first, second, and third constraint solutions, the parameters of the fingerprint image processing model are optimized to obtain a trained fingerprint image processing model.

9. A fingerprint image processing device, characterized in that, include: Image acquisition unit, used to acquire the original image of the contactless fingerprint; The image processing unit is used to process the original image of the non-contact fingerprint using the fingerprint image processing model trained by the model training device as described in claim 8, to obtain a processed non-contact fingerprint image for matching with a contact fingerprint image.

10. An electronic device, characterized in that, include: Memory and processor; The memory is used to store programs; The processor is configured to invoke a program stored in the memory to execute the model training method as described in any one of claims 1 to 6, or to execute the fingerprint image processing method as described in claim 7.

11. A readable storage medium storing a computer program, which, when executed by a processor, implements the model training method as described in any one of claims 1 to 6, or implements the fingerprint image processing method as described in claim 7.

Images