A Deep Learning-Based Fingerprint Ridge Distance Estimation Method and System
By combining spatial and frequency domain information using a lightweight convolutional neural network based on deep learning, a multi-scale fingerprint ridge distance estimation model is constructed, which solves the problems of time-consuming and laborious and insufficient accuracy in existing technologies. It achieves fast and accurate fingerprint ridge distance estimation, which is suitable for scenarios such as mobile devices and robots.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-04
- Publication Date
- 2026-03-06
AI Technical Summary
Existing fingerprint ridge distance estimation methods suffer from problems such as being time-consuming and labor-intensive, having poor adaptability, and insufficient accuracy, especially in applications requiring high image quality or in-situ fingerprinting where performance is significantly limited.
A lightweight convolutional neural network based on deep learning is adopted, which combines spatial and frequency domain information and optimizes the model through mutual learning to construct a multi-scale fingerprint ridge distance estimation model. The lightweight convolutional neural network is then used to achieve fast and accurate estimation of fingerprint ridge distance.
It achieves fast and accurate fingerprint ridge distance estimation, is suitable for local computation in scenarios such as mobile devices and robots, and the model is easy to transfer to applications after training, and can obtain accurate average ridge distance or distribution map of fingerprint images.
Smart Images
Figure CN115690855B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of biometric identification technology, and in particular relates to a fingerprint ridge distance estimation method and system based on deep learning. Background Technology
[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.
[0003] Fingerprint recognition is a technology that identifies individuals by analyzing the papillary ridges (also known as the fingerprint ridges). It is widely used in various fields such as criminal investigation, immigration management, and mobile payments. Although the specific techniques used in different fields vary, they all include three modules: fingerprint acquisition, feature extraction, and fingerprint matching. The average ridge distance of a fingerprint image is one of the important texture attributes of a fingerprint and is often used as an important parameter in calculations and references in some fingerprint segmentation, enhancement, and classification methods. If the measured and estimated ridge distance deviates significantly from the actual distance, the effectiveness and accuracy of the above methods will be greatly affected. Therefore, accurate and refined measurement and estimation of fingerprint ridge distance has significant application value in fingerprint classification, fingerprint image quality assessment, fingerprint enhancement, and fingerprint segmentation methods.
[0004] Existing fingerprint ridge distance estimation methods include:
[0005] (1) Manually marking the distance of fingerprint ridges is time-consuming and labor-intensive, and requires professional fingerprint experts;
[0006] (2) Spatial domain methods: common methods are to count the number of ridges within a fixed window to calculate the local ridge distance, such as estimating the local ridge distance based on the direction of the ridges and the number of ridges within a fixed window; however, they are poorly adaptable to areas with large changes in ridge direction and have high requirements for image quality.
[0007] (3) Frequency domain method: The common method is to transform the fingerprint image into a frequency domain image through Fourier transform, determine the appropriate strong sine wave through the energy distribution in the frequency domain, and use the reciprocal of the frequency of the sine wave as the result of the ridge distance estimation of the image. In some applications, such as on-site fingerprints, the performance of this type of method still has certain limitations. The accuracy of ridge distance estimation still needs to be improved, and it cannot calculate the fingerprint ridge distance distribution in a refined manner.
[0008] Therefore, how to accurately and precisely measure and estimate the distance between fingerprint ridges is a topic worthy of research. Summary of the Invention
[0009] To overcome the shortcomings of the prior art, this invention provides a fingerprint ridge distance estimation method and system based on deep learning. It comprehensively considers spatial and frequency domain information, constructs a multi-scale fingerprint ridge distance estimation model, and further optimizes the model through mutual learning, making the measurement estimation more accurate. Moreover, the trained model is easy to transfer and apply.
[0010] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:
[0011] The first aspect of this invention provides a fingerprint ridge distance estimation method based on deep learning;
[0012] A deep learning-based fingerprint ridge distance estimation method includes:
[0013] The training fingerprint image with labeled ridge distances is cropped into multiple labeled image patches. The labeled image patches with two labeled points that are close in distance and contain their respective ridge distance labels are grouped together to construct the training dataset.
[0014] A fingerprint ridge distance estimation model based on a lightweight convolutional neural network is constructed. The labeled image patches are input into two fingerprint ridge distance estimation models in pairs according to the group. The structural similarity loss of the two labeled image patches in the same group promotes mutual learning between the two models. The models are trained in a distributed manner. When the model performance converges, the training is stopped, and the model with the best performance among the two models is selected as the final fingerprint ridge distance estimation model.
[0015] The fingerprint image to be estimated is cropped into multiple image blocks, and each image block is input into the final fingerprint ridge distance estimation model. Based on the predicted ridge distance estimate of the individual image block sample, the fingerprint ridge distance of the fingerprint image is calculated.
[0016] Furthermore, experts manually annotated the ridge line distances, and then extracted a 64×64 neighboring image block centered on the annotated location.
[0017] The annotation sites within a group should be located close to each other, while the annotation sites between groups should be selected from different fingerprint regions.
[0018] Furthermore, the fingerprint ridge distance estimation model consists of convolutional layers, pooling layers, and fully connected layers;
[0019] The radial energy spectrum in the frequency domain of the fingerprint image block is input into a fully connected layer and connected with the feature map extracted from the image block to obtain the feature vector of the image block.
[0020] Furthermore, the feature vector is decoded into fingerprint ridge distance by a numerical decoder; the fingerprint ridge distance is encoded into feature vector by a numerical encoder.
[0021] Furthermore, during the training of a single model, the target feature vector, which is encoded by a numerical encoder, corresponding to the ridge distance value of the image patch, is compared with the feature vector output by the model to calculate the loss through mean square error, thereby training and optimizing the model.
[0022] Furthermore, the structural similarity between two labeled image blocks in the same group means that the labeled locations of the image blocks in the group are similar, the ridge structure information of the two image blocks in the group is similar, and the ridge distance values are similar.
[0023] Furthermore, the ridge distance estimates of multiple image blocks are calculated using weighted averages to obtain the fingerprint ridge distance of the fingerprint image.
[0024] A second aspect of the present invention provides a fingerprint ridge distance estimation system based on deep learning.
[0025] A fingerprint ridge distance estimation system based on deep learning includes a data construction module, a model training module, and a distance prediction module.
[0026] The data construction module is configured to: extract the training fingerprint image with labeled ridge distances into multiple labeled image patches, and group the labeled image patches with two labeled points that are close to each other and contain their respective ridge distance labeled values to construct the training dataset;
[0027] The model training module is configured to: construct a fingerprint ridge distance estimation model based on a lightweight convolutional neural network; input the labeled image patches in pairs into two fingerprint ridge distance estimation models according to groups; use the structural similarity loss of the two labeled image patches in the same group to promote mutual learning between the two models; perform distributed training on the models; stop training when the model performance converges; and select the model with the best performance among the two models as the final fingerprint ridge distance estimation model.
[0028] The distance prediction module is configured to: crop the fingerprint image to be estimated into multiple image blocks, input a single image block into the final fingerprint ridge distance estimation model, and calculate the fingerprint ridge distance of the fingerprint image based on the predicted ridge distance estimate of the single image block sample.
[0029] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of a deep learning-based fingerprint ridge distance estimation method as described in the first aspect of the present invention.
[0030] A fourth aspect of the present invention provides an electronic device including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a deep learning-based fingerprint ridge distance estimation method as described in the first aspect of the present invention.
[0031] The above one or more technical solutions have the following beneficial effects:
[0032] This invention can quickly and accurately estimate fingerprint ridge distance using deep learning technology, without requiring extensive knowledge of fingerprint recognition. The model measures and estimates fingerprint ridge distance at multiple scales, comprehensively considering spatial and frequency domain information in its framework. Furthermore, the model is optimized through mutual learning, making the measurement estimation more accurate, and the trained model is easy to transfer and apply.
[0033] This invention is a lightweight convolutional neural network that can quickly measure and estimate the ridge distance of fingerprint images. It is suitable for local computation in mobile devices, robots, etc., and facilitates applications in fingerprint recognition, repair, and other scenarios. The model proposed in this invention consists of two stages: training and prediction. The computational complexity of the model is relatively higher in the training stage, but the model parameter scale in the prediction stage after training is much smaller than that of classic network structures such as VGG. Moreover, different input scales can be set in the prediction stage according to application requirements to obtain more accurate average ridge distance values or ridge distance distribution maps of fingerprint images.
[0034] This invention can be further applied to solve more complex problems of the same kind, such as palm print ridge distance estimation and texture feature recognition, and is scalable and highly applicable.
[0035] Advantages of additional aspects 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
[0036] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0037] Figure 1 This is a schematic diagram of the fingerprint ridge distance estimation machine learning model in the first embodiment.
[0038] Figure 2 This is a flowchart of the method in the first embodiment.
[0039] Figure 3 This is a structural diagram of the fingerprint image ridge distance estimation model for the first embodiment.
[0040] Figure 4 This is a diagram illustrating the prediction process of the fingerprint image ridge distance estimation model in the first embodiment.
[0041] Figure 5 This is a system structure diagram of the second embodiment. Detailed Implementation
[0042] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0043] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the invention; unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0044] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention; as used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise; furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.
[0045] Example 1
[0046] This embodiment discloses a fingerprint ridge distance estimation method based on deep learning, such as... Figure 1 As shown, the process mainly consists of three stages: construction, training, and prediction. After training with the constructed dataset, the resulting model can be used for prediction over a long period of time. When the data accumulates to a certain scale, the model can be retrained, and then the training results (model description parameters) can be updated. Although the training process is computationally intensive, the lightweight design of the model itself does not require large-scale storage. Therefore, the trained model can be easily transferred and applied, providing more accurate ridge distance information for application scenarios such as fingerprint recognition and fingerprint image enhancement.
[0047] The training and prediction processes of fingerprint ridge distance estimation models are basically similar, with the following differences: During training, image patches with similar labeled points are selected and input into two models in pairs, and the ridge distance labels of the two image patches are input into the numerical encoder as target feature vectors for model training; while during prediction, only fingerprint image patches are input, and after the feature vectors are output, the estimated ridge distance of the image patch is output through the numerical decoder.
[0048] like Figure 2 As shown, a fingerprint ridge distance estimation method based on deep learning specifically includes:
[0049] Step S1: Extract the training fingerprint image with labeled ridge distances into multiple labeled image patches. Group the labeled image patches with two labeled points that are close to each other and contain their respective ridge distance labels to construct the training dataset.
[0050] Furthermore, experts manually annotated the ridge line distances, and then extracted a 64×64 neighboring image block centered on the annotated location.
[0051] The annotation sites within a group should be located close to each other, while the annotation sites between groups should be selected from different fingerprint regions.
[0052] The fingerprint images used for training are first manually labeled with ridge distances by experts. Then, 64×64 neighboring image blocks are cropped with the labeled location as the center. Two image blocks with close distances between labeled locations are grouped together and contain their respective ridge distance labels. This is used as the training dataset.
[0053] The training fingerprint images used in this embodiment contain hundreds of fingerprints, and each fingerprint is labeled at least 6 times, with every two labels forming a group. During the labeling process, the labeled sites within a group are required to be close in location, while the labeled sites between groups are selected from different fingerprint regions to perform a more accurate and comprehensive ridge distance estimation.
[0054] Different fingerprint regions are different locations and different parts, requiring that the two regions be far apart; the locations of the annotation sites between groups need to be relatively far apart in order to collect fingerprint ridge distance information of different local areas, so as to make the annotation results more comprehensive.
[0055] Step S2: Construct a fingerprint ridge distance estimation model based on a lightweight convolutional neural network. Input the labeled image patches in pairs into two fingerprint ridge distance estimation models according to the group. The structural similarity loss of the two labeled image patches in the same group promotes mutual learning between the two models. Perform distributed training on the models. When the model performance converges, stop training and select the model with the best performance among the two models as the final fingerprint ridge distance estimation model.
[0056] The fingerprint ridge distance estimation model used in this embodiment is essentially a lightweight convolutional neural network. Convolutional neural networks (CNNs) are an important component of machine learning, generally consisting of several convolutional layers, pooling layers, and fully connected layers. Convolutional layers are used for image feature extraction, pooling layers are used for sampling the features extracted by the convolutional layers, and fully connected layers are used to connect the features extracted by the convolutional network. CNNs can automatically learn local feature extractors and implement the weight sharing principle, thereby reducing the number of free parameters and improving performance compared to neural network architectures, demonstrating great achievements in the field of image recognition.
[0057] The structure of the fingerprint ridge distance estimation model is as follows: Figure 3As shown, multiple non-overlapping 64x64 pixel fingerprint image patches, cropped from the largest window region of the input fingerprint image, are used as input to the model. The ridge distance value of each image patch is automatically calculated. Through the fingerprint ridge distance estimation model, a more refined fingerprint ridge distance distribution can be obtained, and the ridge distance values of multiple image patches in the image can be weighted to calculate the fingerprint ridge distance more accurately. The largest window region is the largest rectangular window region within the fingerprint region of interest.
[0058] From the perspective of model structure, the fingerprint ridge distance estimation model is improved by inputting the radial energy spectrum in the frequency domain of the fingerprint image block into the fully connected layer and connecting it with the feature map extracted from the image block to obtain the feature vector of the image block. This increases the model's reference to the image frequency domain information and improves the reference range of the model information.
[0059] During training, labeled image patches are input in pairs according to groups to perform distributed training of two fingerprint ridge distance estimation models. Since the labeled sites of image patches within the same group are close, the ridge structure information of the two image patches within the same group is similar and the ridge distance values are similar. In this way, structural similarity loss is set to promote mutual learning between the two networks.
[0060] Finally, the ridge distance values corresponding to the image patches are encoded into vectors by a numerical encoder and used as target feature vectors. These vectors are then compared with the model's output feature vectors to calculate the mean squared error loss, thus training and optimizing the model. During iterative training, training stops when the model performance converges, and the best-performing model among the two trained models is selected as the final fingerprint ridge distance estimation model. To improve the ridge distance estimation accuracy on the validation set, the model parameters can be adjusted and optimized. The training set is used for model training, and the validation set is used to evaluate the model. Based on the model's prediction and estimation performance on the validation set, the model's hyperparameters are adjusted and optimized.
[0061] Step S3: The fingerprint image to be estimated is cropped into multiple image blocks, and each image block is input into the final fingerprint ridge distance estimation model. Based on the predicted ridge distance estimate of the single image block sample, the fingerprint ridge distance of the fingerprint image is calculated.
[0062] like Figure 4 As shown, the fingerprint image to be estimated is split and cropped into multiple image patch samples as needed, and a single image patch sample is input into the trained model. The feature vector output by the model is fed into the input numerical decoder to obtain the ridge distance estimate of the image patch, and the result is a specific numerical value. According to the application requirements, the ridge distance estimates of multiple image patches can be calculated by weighting to obtain a more accurate fingerprint image ridge distance, or the input fingerprint image can be split into non-overlapping image patches and input into the model to draw a fingerprint image ridge distance distribution map.
[0063] On-demand splitting involves setting different sampling parameters based on different application scenarios and usage requirements. In scenarios with poor hardware conditions or low accuracy requirements, the number of image block samples extracted by cropping is relatively small, such as 3 or 5, and the model measurement speed is also relatively faster, only obtaining the ridge distance estimate of the entire fingerprint image.
[0064] In application scenarios with complete hardware conditions or high accuracy requirements, more refined cropping can be performed, and a larger number of image block samples can be collected to perform more refined fingerprint ridge distance estimation or obtain fingerprint image ridge distance distribution map.
[0065] Example 2
[0066] This embodiment discloses a fingerprint ridge distance estimation system based on deep learning;
[0067] like Figure 5 As shown, a fingerprint ridge distance estimation system based on deep learning includes a data construction module, a model training module, and a distance prediction module.
[0068] The data construction module is configured to: extract the training fingerprint image with labeled ridge distances into multiple labeled image patches, and group the labeled image patches with two labeled points that are close to each other and contain their respective ridge distance labeled values to construct the training dataset;
[0069] The model training module is configured to: construct a fingerprint ridge distance estimation model based on a lightweight convolutional neural network; input the labeled image patches in pairs into two fingerprint ridge distance estimation models according to groups; use the structural similarity loss of the two labeled image patches in the same group to promote mutual learning between the two models; perform distributed training on the models; stop training when the model performance converges; and select the model with the best performance among the two models as the final fingerprint ridge distance estimation model.
[0070] The distance prediction module is configured to: crop the fingerprint image to be estimated into multiple image blocks, input a single image block into the final fingerprint ridge distance estimation model, and calculate the fingerprint ridge distance of the fingerprint image based on the predicted ridge distance estimate of the single image block sample.
[0071] Example 3
[0072] The purpose of this embodiment is to provide a computer-readable storage medium.
[0073] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of a deep learning-based fingerprint ridge distance estimation method as described in Embodiment 1 of this disclosure.
[0074] Example 4
[0075] The purpose of this embodiment is to provide an electronic device.
[0076] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in a deep learning-based fingerprint ridge distance estimation method as described in Embodiment 1 of this disclosure.
[0077] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A deep learning-based fingerprint ridge distance estimation method, characterized in that, The method comprises the following steps: Cut the training fingerprint image with labeled ridge distance into multiple labeled image blocks, and take two labeled image blocks with similar labeled point positions and containing respective ridge distance labels as a group to construct a training data set; The labeled point positions in the same group are similar, and the labeled point positions in different groups are selected from different fingerprint regions; Construct a fingerprint ridge distance estimation model based on a lightweight convolutional neural network, input the labeled image blocks in pairs into two fingerprint ridge distance estimation models according to the grouping, the structural similarity loss of the two labeled image blocks in the same group promotes the mutual learning of the two models, and the models are distributedly trained, the training is stopped when the model performance converges, and the model with the best performance in the two models is selected as the final fingerprint ridge distance estimation model; Cut the fingerprint image to be estimated into multiple image blocks, input a single image block into the final fingerprint ridge distance estimation model, and calculate the fingerprint ridge distance of the fingerprint image based on the predicted ridge distance estimation value of the single image block sample; Multiple non-overlapping 64x64 pixel size fingerprint image blocks cut from the maximum window region of the input fingerprint image are taken as the input of the model, the ridge distance values of the image blocks are automatically calculated, the fingerprint ridge distance distribution is obtained through the fingerprint ridge distance estimation model, and the ridge distance values of multiple image blocks in the image are weighted to more accurately measure the fingerprint ridge distance; The feature vector is decoded into a fingerprint ridge distance by a numerical decoder; and the fingerprint ridge distance is encoded into a feature vector by a numerical encoder.
2. The deep learning-based fingerprint ridge distance estimation method of claim 1, wherein, The fingerprint ridge distance estimation model is composed of a convolutional layer, a pooling layer and a fully connected layer; The radial energy spectrum in the frequency domain of the fingerprint image block is input into the fully connected layer, and the feature map extracted from the image block is connected to obtain the feature vector of the image block.
3. The deep learning-based fingerprint ridge distance estimation method of claim 1, wherein, In the training of a single model, the target feature vector obtained by encoding the ridge distance value of the image block through the numerical encoder is compared with the feature vector output by the model to calculate the mean square error loss, so as to train and optimize the model.
4. The deep learning-based fingerprint ridge distance estimation method of claim 1, wherein, The structural similarity of the two labeled image blocks in the same group is that the labeled point positions in the same group are similar, and the ridge structure information of the two image blocks in the same group is similar and the ridge distance values are similar.
5. The deep learning based fingerprint ridge distance estimation method of claim 1, wherein, The ridge distance estimation values of multiple image blocks are weighted to obtain the fingerprint ridge distance of the fingerprint image. 6.A deep learning based fingerprint ridge distance estimation system, characterized in that, The method comprises a data construction module, a model training module and a distance prediction module: The data construction module is configured to cut the training fingerprint image with labeled ridge distance into multiple labeled image blocks, take two labeled image blocks with similar labeled point positions and containing respective ridge distance labels as a group to construct a training data set, and manually label the ridge distance by experts, and cut a 64x64 neighborhood image block with the labeled point position as the center; The labeled point positions in the same group are similar, and the labeled point positions in different groups are selected from different fingerprint regions; The model training module is configured to: construct a fingerprint ridge distance estimation model based on a lightweight convolutional neural network, input a pair of labeled image blocks in a group to two fingerprint ridge distance estimation models, promote mutual learning of the two models by structural similarity loss of the two labeled image blocks in the same group, perform distributed training on the models, stop training when the model performance converges, and select the model with the best performance from the two models as the final fingerprint ridge distance estimation model; The distance prediction module is configured to: cut the fingerprint image to be estimated into multiple image blocks, input a single image block into the final fingerprint ridge distance estimation model, and calculate the fingerprint ridge distance of the fingerprint image based on the predicted ridge distance estimation value of the single image block sample. Multiple non-overlapping 64x64 pixel size fingerprint image blocks cut from the maximum window area of the input fingerprint image are used as the input of the model, the ridge distance values of the image blocks are automatically calculated, a more refined fingerprint ridge distance distribution is obtained through the fingerprint ridge distance estimation model, and the ridge distance values of multiple image blocks in the image can be weighted to more accurately measure the fingerprint ridge distance. The feature vector is decoded into a fingerprint ridge distance by a numerical decoder, and the fingerprint ridge distance is encoded into a feature vector by a numerical encoder.
7. A computer-readable storage medium having stored thereon a program, characterized in that The program is executed by the processor to implement the steps of the fingerprint ridge distance estimation method based on deep learning in any one of claims 1-5.
8. An electronic device comprising a memory, a processor, and a program stored in the memory and capable of running on the processor, characterized by The processor executes the program to implement the steps of the fingerprint ridge distance estimation method based on deep learning in any one of claims 1-5.