A fingerprint texture extraction method, system, device and storage medium based on deep learning

By employing a deep learning-based fingerprint texture extraction method, utilizing an encoder-decoder structure and a feature fusion layer in the ERFNet model, the difficulty of texture extraction in the case of damaged fingerprints by traditional methods is solved, thereby improving the fingerprint recognition accuracy.

CN113239808BActive Publication Date: 2024-11-19GRG BANKING EQUIPMENT CO LTD
View PDF 9 Cites 0 Cited by

Patent Information

Application Number
CN202110528813.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-05-14
Publication Date
2024-11-19
Estimated Expiration
2041-05-14

AI Technical Summary

Technical Problem

Traditional fingerprint recognition methods struggle to effectively extract textures when fingerprints are blurry, damaged, or smudged, resulting in low recognition accuracy.

Method used

A deep learning-based fingerprint texture extraction method is adopted, which utilizes an encoder-decoder segmentation network model combined with a feature fusion layer to preprocess, extract, and crop fingerprint images, including grayscale processing, pixel filling, and normalization. The ERFNet semantic segmentation model is used for texture segmentation.

Benefits of technology

It improves the accuracy of fingerprint recognition, effectively extracts fingerprint textures under various damage conditions, and enhances the model's texture extraction capability and adaptability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113239808B_ABST
    Figure CN113239808B_ABST
Patent Text Reader

Abstract

The present invention discloses a fingerprint texture extraction method, system, device and storage medium based on deep learning, wherein the fingerprint texture extraction method specifically includes step S1: obtaining a fingerprint image, and preprocessing the fingerprint image so that the fingerprint image size meets the input requirement of the fingerprint texture extraction model; step S2: inputting the preprocessed fingerprint image into the fingerprint texture extraction model to output a texture map; step S3: cropping the output texture map to obtain a texture map that matches the size of the original fingerprint image. The method can well extract fingerprint textures in various situations, and can greatly improve the accuracy of fingerprint recognition.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of fingerprint recognition technology, and in particular to a fingerprint texture extraction method, system, device and storage medium based on deep learning. Background Art

[0002] With the continuous development of the information society, people are in urgent need of more reliable identification technology to authenticate their identities. Using biometrics for identity authentication has become a trend nowadays. Among them, fingerprint recognition is widely recognized and applied due to its convenience and reliability. The traditional fingerprint recognition method is to first extract the ridge texture map of the fingerprint, then extract feature points based on the texture map and match them to find similarities. Among them, fingerprint texture extraction is a very important link, which directly affects the accuracy of subsequent feature point extraction and matching.

[0003] However, although traditional methods can achieve high recognition rates on high-quality fingerprint images, in actual applications, the fingerprints may be blurred, damaged, or contaminated during the fingerprint collection process, making it difficult to extract complete fingerprint textures based on traditional methods, resulting in poor fingerprint recognition results. In particular, when fingerprints are severely damaged, traditional methods are unable to restore fingerprint textures, making it impossible to improve fingerprint recognition accuracy. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, one of the objectives of the present invention is to provide a fingerprint texture extraction method based on deep learning, which can improve the accuracy of fingerprint recognition.

[0005] A second object of the present invention is to provide a fingerprint texture extraction system based on deep learning.

[0006] A third object of the present invention is to provide a fingerprint texture extraction device.

[0007] A fourth object of the present invention is to provide a storage medium.

[0008] One of the purposes of the present invention is achieved by the following technical solution:

[0009] A fingerprint texture extraction method based on deep learning, comprising:

[0010] Step S1: Acquire a fingerprint image and preprocess the fingerprint image so that the fingerprint image size meets the input requirement of the fingerprint texture extraction model;

[0011] Step S2: inputting the preprocessed fingerprint image into the fingerprint texture extraction model to output a texture map;

[0012] Step S3: Crop the output texture image to obtain a texture image that matches the size of the original fingerprint image.

[0013] Furthermore, the method for preprocessing the fingerprint image is:

[0014] The collected fingerprint image is gray-processed, and then the pixels around the fingerprint image are filled according to the preset image ratio and size;

[0015] Normalize the pixel values ​​of the image after filling the pixels.

[0016] Furthermore, the preset image size is that both the length and width of the image are multiples of 8, so that the width and height of the filled image are:

[0017]

[0018] in, is the rounding symbol, W is the width of the padded image, H is the height of the padded image, w is the width of the original image, and h is the height of the original image.

[0019] Furthermore, the pixel value of the pixels used to fill the periphery of the image is 128, and the number of the filling pixels is:

[0020]

[0021] in, The floor symbol.

[0022] Furthermore, the formula for normalizing the pixel values ​​of the image after filling the pixels is:

[0023] F(x,y)=f(x,y) / 255-0.5 (Formula 3)

[0024] Among them, f(x,y) is the pixel value before normalization, F(x,y) is the pixel value after normalization, and F(x,y)∈[-0.5,0.5].

[0025] Furthermore, the fingerprint texture extraction model adopts a segmentation network model of an encoder-decoder structure, and a feature fusion layer is provided between the weaving part and the decoder part for fusing the features of the encoder and the decoder.

[0026] Furthermore, the encoder comprises:

[0027] Multiple downsampling layers for downsampling images;

[0028] Multiple NB1D layers are used to perform convolution processing on the downsampled image to extract image features, wherein the NB1D layer uses 3*1 and 1*3 grouped convolutions.

[0029] The second object of the present invention is achieved by adopting the following technical solution:

[0030] A fingerprint texture extraction system based on deep learning, comprising:

[0031] A preprocessing module is used to preprocess the acquired fingerprint image so that the fingerprint image size meets the input requirement of the fingerprint texture extraction model;

[0032] A model analysis module, used for inputting the pre-processed fingerprint image into the fingerprint texture extraction model to output a texture map;

[0033] The post-processing module is used to crop the output texture image to obtain a texture image that matches the size of the original fingerprint image.

[0034] The third object of the present invention is achieved by adopting the following technical solution:

[0035] A fingerprint texture extraction device, comprising:

[0036] program;

[0037] A memory, used for storing the program;

[0038] A processor is used to load the program to execute the fingerprint texture extraction method based on deep learning as described above.

[0039] The fourth object of the present invention is achieved by adopting the following technical solution:

[0040] A storage medium stores a program, characterized in that when the program is executed by a processor, the fingerprint texture extraction method based on deep learning as described above is implemented.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The present invention provides a method based on deep learning to extract fingerprint texture images. For the collected fingerprint images, the fingerprint texture images are directly extracted using the proposed semantic segmentation model. The method utilizes the powerful learning ability of deep learning, so that the model has good texture extraction ability and adaptability to various damage conditions, thereby improving the accuracy of fingerprint texture extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the flow of the fingerprint texture extraction method based on deep learning of the present invention;

[0044] Figure 2 It is a structural schematic diagram of the fingerprint texture extraction model of the present invention;

[0045] Figure 3 Schematic diagram of the structure of the downsampling layer of the present invention;

[0046] Figure 4It is a structural schematic diagram of the characteristic fusion layer of the present invention;

[0047] Figure 5 It is a schematic diagram of the structure of the NB1D layer of the present invention;

[0048] Figure 6 It is a schematic diagram of the structure of the NB1D_Dx layer of the present invention;

[0049] Figure 7 This is a fingerprint texture extraction effect diagram of the present invention;

[0050] Figure 8 This is a schematic block diagram of the modules of the fingerprint texture extraction system based on deep learning of the present invention. DETAILED DESCRIPTION

[0051] The present invention is further described below in conjunction with the accompanying drawings and specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form a new embodiment.

[0052] Embodiment 1

[0053] This embodiment provides a fingerprint texture extraction method based on deep learning. The method of this embodiment can well extract fingerprint textures in various situations and greatly improve the accuracy of fingerprint recognition.

[0054] like Figure 1 As shown, the fingerprint texture extraction method of this embodiment specifically includes the following steps:

[0055] Step S1: Acquire a fingerprint image and preprocess the fingerprint image so that the fingerprint image size meets the input requirement of the fingerprint texture extraction model;

[0056] Step S2: inputting the preprocessed fingerprint image into the fingerprint texture extraction model to output a texture map;

[0057] Step S3: Crop the output texture image to obtain a texture image that matches the size of the original fingerprint image.

[0058] Among them, the fingerprint image can be obtained through the Internet or a fingerprint collection device. Since the size and proportion of the collected fingerprint image may be uneven, the collected fingerprint image needs to be preprocessed. In this embodiment, the preprocessing step of the fingerprint image mainly includes two steps. One is to grayscale the collected fingerprint image so that the fingerprint image is unified into a fingerprint grayscale image, and then fill the pixels around the fingerprint image according to the preset image proportion and size; in this embodiment, pixels are filled around the fingerprint image, and the size and proportion of the fingerprint image after the pixels are filled can meet the preset requirements; in this embodiment, the pixel value filled in the fingerprint image is unified to 128 pixels, and it is ensured that the width and height of the fingerprint image after the pixels are filled are multiples of 8, where the original width of the image is w and the original length is h, then the width W and height H of the image after the pixels are filled are as shown in Formula 1:

[0059]

[0060] in, This ensures that the width W and height H of the transformed image are multiples of 8 and are slightly larger integers closest to the original width w and height h.

[0061] It is obvious that

[0062] (Ww)∈[0,1,...,7], (Hh)∈[0,1,...,7]

[0063] Then, the number of pixels filled in the top, bottom, left and right of the transformed image is:

[0064]

[0065] in, This transformation tries to ensure the aspect ratio and size of the original image, so that the fingerprint input image does not deform, and also meets the requirement of the texture extraction model that the input image size is an integer multiple of 8.

[0066] The fingerprint image preprocessing also includes the second step: normalizing the pixel values ​​of the image after filling the pixels. The normalization process divides each pixel by 255 and then subtracts 0.5, which can be expressed as:

[0067] F(x,y)=f(x,y) / 255-0.5 (Formula 3)

[0068] Among them, f(x,y) is the pixel value before normalization, F(x,y) is the pixel value after normalization, and F(x,y)∈[-0.5,0.5].

[0069] After the collected fingerprint image is pixel-filled and normalized, the size and specifications of the fingerprint image can be unified, which facilitates the subsequent input of the image into the fingerprint texture extraction model for texture extraction.

[0070] After the fingerprint image is preprocessed, it is imported into the fingerprint texture extraction model, where the fingerprint texture extraction model is an improvement based on the ERFNet semantic segmentation model to adapt to the segmentation and extraction of fingerprint texture.

[0071] The fingerprint texture extraction model of this embodiment adopts an encoder-decoder segmentation network structure, in which the encoder is the encoder part, which is used to gradually reduce the spatial dimension of the input data; the decoder is the decoder part, which is used to gradually restore the details of the target and the corresponding spatial dimension; and a feature fusion layer (merge layer) is provided between the encoder and the decoder, which is used to fuse the features of the encoder and the decoder to fuse the shallow and deep features. The network framework of the fingerprint texture extraction model of this embodiment is as follows Figure 2 As shown, the left side of the figure is the encoder part and the right side is the decoder part.

[0072] The encoder part includes stacked downsampling layers, NB1D layers, and NB1D-Dx layers.

[0073] For the semantic segmentation task, it is essential to downsample the image. The downsampling layer Down in this embodiment uses the downsampling module Downsamplerblock in ERFNet. The 2x, 4x, and 8x in front of it represent that the current downsampled output feature map size is 1 / 2, 1 / 4, and 1 / 8 of the original input size, which can significantly reduce the feature map size, reduce the network calculation complexity and memory occupancy rate, and enhance the real-time performance of segmentation. And because the size of the initial input image is large and contains more redundant information, resulting in high memory occupancy rate and calculation complexity, affecting the running speed of the network model, therefore, in this embodiment, the input fingerprint image will be downsampled once to halve the input image size, reduce a large amount of redundancy of the visual information itself, save a lot of calculations, and further improve the running speed of the network model.

[0074] The structure of the lower acquisition layer in this embodiment is as follows Figure 3As shown in the figure, using MaxPooling of size=2*2, stride=2 and convolution kernel of filter=3*3, stride=2, they are finally combined as the down-sampled output. For example, the input image of size 1024*512*3 is convolved by 3*3, and the convolution step size stride is 2 to obtain 13 channels. The right side is convolved by MaxPooling to obtain 3 channels, which are merged by concat operation to obtain 16 channels, so that the final image becomes 512*256*16 in size.

[0075] In this embodiment, the input image is processed by the NB1D layer after a layer of downsampling. The NB1D layer in this embodiment is based on Non-bottleneck-1D (Non-bt-1D) with BatchNorm and dropout layers added. The structure of the NB1D layer is as follows: Figure 5 As shown in the figure, two 3*3 convolution kernels are decomposed into two groups of 3*1 and 1*3 one-dimensional convolutions to reduce the amount of calculation; and after each group of 3*1 and 1*3 one-dimensional convolutions is completed, the BatchNorm layer is required to perform BN normalization processing to forcibly pull the distribution of the output value of any neuron in each layer of the neural network back to the standard normal distribution with a mean of 0 and a variance of 1. Finally, the dropout layer using the discard method is used to reduce overfitting.

[0076] In this embodiment, the data after a downsampling operation passes through 2 NB1D layers, then 1 downsampling layer, 5 NB1D layers and 1 downsampling layer, and then enters the NB1D-Dx layer; in this embodiment, the NB1D-Dx layer is stacked with NB1D-D2 layer, NB1D-D4 layer, NB1D-D8 layer and NB1D-D16 layer in sequence. Among them, the NB1D-Dx layer is based on the dilated convolution module of the NB1D layer, where the dilated convolution is also called the expanded convolution, which uses dilated convolution on the last two layers of NB1D convolution, where the x in dx represents the dilated value (expansion coefficient); the structure of the NB1D-Dx layer in this embodiment is as follows Figure 6 As shown in the figure, the NB1D-Dx layer can not only keep the image spatial information from being lost, but also increase the local receptive field of the network, thereby improving the semantic segmentation accuracy.

[0077] After the fingerprint image is processed by the encoder, it enters the decoder part for upsampling and convolution operations to complete the fingerprint pattern extraction operation. The encoder in this embodiment includes stacked NB1D-Dx layers, upper acquisition layers and NB1D layers, wherein the structure of each layer in the decoder is the same as the corresponding layer structure in the encoder, except that the stacking order of the layers in the decoder is opposite to that of the encoder; in this embodiment, after the image data after the encoder enters the decoder, it passes through the NB1D-D2 layer, NB1D-D4 layer, NB1D-D8 layer and NB1D-D16 layer in sequence, and then passes through 1 upsampling layer, 2 NB1D layers, 1 upsampling layer, 2 NB1D layers and 1 upsampling layer. The final model outputs a fingerprint texture image with the same size W*H as the input. And during the decoding process, the Merge feature fusion module is used to fuse the features in the corresponding feature maps between the encoder and the decoder to fuse the shallow and deep features to better fit the fingerprint texture. In this embodiment, the merge layer is as follows Figure 4 As shown in the figure, the features of the encoder and decoder are merged and then a 1*1 convolution is performed.

[0078] After the fingerprint texture image is extracted by the fingerprint texture extraction model in this embodiment, the output texture map size is W×H. It is necessary to obtain a texture map of the same size as the original image by cropping the pixels filled in the top, bottom, left and right of the image during image preprocessing. The top, bottom, left and right values ​​to be cropped are calculated by formula 2.

[0079] The quality of fingerprint texture extraction is directly related to the extraction and matching of fingerprint feature points, which is the most important link in fingerprint recognition. In the present invention, a deep learning model - fingerprint texture extraction model is used to directly extract fingerprint texture images from the collected fingerprint images, such as Figure 7 As shown in the figure, the fingerprint texture can be extracted well in various situations, solving the problem of fingerprint texture extraction in the case of blur, damage, and severe contamination, which can directly improve the accuracy of fingerprint recognition. This method uses the powerful learning ability of deep learning, so that the model has good texture extraction ability and adaptability to various damage situations.

[0080] Embodiment 2

[0081] This embodiment provides a fingerprint texture extraction system based on deep learning, which executes the fingerprint texture extraction method based on deep learning described in Embodiment 1, such as Figure 8 As shown, the fingerprint texture extraction system of this embodiment includes at least the following modules:

[0082] A preprocessing module is used to preprocess the acquired fingerprint image so that the fingerprint image size meets the input requirement of the fingerprint texture extraction model;

[0083] A model analysis module, used for inputting the pre-processed fingerprint image into the fingerprint texture extraction model to output a texture map;

[0084] The post-processing module is used to crop the output texture image to obtain a texture image that matches the size of the original fingerprint image.

[0085] Embodiment 3

[0086] This embodiment provides a fingerprint texture extraction device, including:

[0087] program;

[0088] A memory, used for storing the program;

[0089] A processor is used to load the program to execute the fingerprint texture extraction method based on deep learning described in Embodiment 1.

[0090] In addition, this embodiment also provides a storage medium storing a program, characterized in that when the program is executed by a processor, the fingerprint texture extraction method based on deep learning as described above is implemented.

[0091] The device and storage medium in this embodiment and the method in the above embodiment are two aspects based on the same inventive concept. The method implementation process has been described in detail above, so those skilled in the art can clearly understand the structure and implementation process of the device in this implementation based on the above description. For the sake of brevity of the specification, they will not be repeated here. The above implementation is only a preferred implementation of the present invention, and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and replacements made by those skilled in the art on the basis of the present invention belong to the scope of protection claimed by the present invention.

[0092] The above-mentioned embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and substitutions made by technicians in this field on the basis of the present invention shall fall within the scope of protection required by the present invention.

Claims

1. A fingerprint texture extraction method based on deep learning, characterized in that: include: Step S1: Acquire a fingerprint image and preprocess the fingerprint image so that the fingerprint image size meets the input requirement of the fingerprint texture extraction model; Step S2: Input the preprocessed fingerprint image into the fingerprint texture extraction model to output a texture map; wherein the fingerprint texture extraction model adopts a segmentation network model of an encoder-decoder structure, and a feature fusion layer is provided between the encoder part and the decoder part, which is used to fuse the features of the encoder and the decoder; wherein the encoder includes: multiple downsampling layers for image downsampling processing; multiple NB1D layers, the NB1D layer is based on Non-bottleneck-1D with BatchNorm and dropout layers added, and the NB1D layer adopts two groups of 3 *1, 1*3 one-dimensional convolution, and after each set of 3*1, 1*3 one-dimensional convolution, the BatchNorm layer performs BN normalization processing, which is used to regress the distribution of the output value of any neuron in each layer of the neural network to a standard normal distribution with a mean of 0 and a variance of 1, and then passes through a dropout layer using the discard method to reduce overfitting, and implements convolution processing on the downsampled image to extract image features; the NB1D-Dx layer is a dilated convolution module based on the NB1D layer, that is, dilated convolution is used on the last two layers of NB1D convolution, where the x in Dx represents the dilation coefficient; Step S3: Crop the output texture image to obtain a texture image that matches the size of the original fingerprint image.

2. The fingerprint texture extraction method based on deep learning according to claim 1 is characterized in that: The method for preprocessing the fingerprint image is: grayscale processing the collected fingerprint image, and then filling pixels around the fingerprint image according to a preset image ratio and size; Normalize the pixel values ​​of the image after filling the pixels.

3. The fingerprint texture extraction method based on deep learning according to claim 2 is characterized in that: The preset image size is that the length and width of the image are both multiples of 8, so that the width and height of the padded image are: where is the rounding symbol, W is the width of the padded image, H is the height of the padded image, w is the width of the original image, and h is the height of the original image.

4. The fingerprint texture extraction method based on deep learning according to claim 3 is characterized in that: The pixel value of the pixels used to fill the four sides of the image is 128, and the number of filling pixels is: wherein is a rounding-down symbol.

5. The fingerprint texture extraction method based on deep learning according to claim 2 is characterized in that: The formula for normalizing the pixel value of the image after filling pixels is: F(x,y)=f(x,y) / 255-0.5, wherein f(x,y) is the pixel value before normalization, F(x,y) is the pixel value after normalization, and F(x,y)∈[-0.5,0.5].

6. A fingerprint texture extraction system based on deep learning, characterized in that: include: A preprocessing module is used to preprocess the acquired fingerprint image so that the fingerprint image size meets the input requirement of the fingerprint texture extraction model; The model analysis module is used to input the pre-processed fingerprint image into the fingerprint texture extraction model to output a texture map; wherein the fingerprint texture extraction model adopts a segmentation network model of an encoder-decoder structure, and a feature fusion layer is provided between the weaving part and the decoder part, which is used to fuse the features of the encoder and the decoder; wherein the encoder includes: multiple downsampling layers for processing the image; multiple NB1D layers, the NB1D layer is based on Non-bottleneck-1D with BatchNorm and dropout layers, and the NB1D layer adopts two A group of 3*1 and 1*3 one-dimensional convolutions is performed, and after each group of 3*1 and 1*3 one-dimensional convolutions is completed, a BatchNorm layer is used to perform BN normalization processing, which is used to regress the distribution of the output value of any neuron in each layer of the neural network to a standard normal distribution with a mean of 0 and a variance of 1. Then, a dropout layer using the discarding method is used to reduce overfitting, and convolution processing is performed on the downsampled image to extract image features; the NB1D-Dx layer is a dilated convolution module based on the NB1D layer, that is, dilated convolution is used on the last two layers of convolution of NB1D, where the x in Dx represents the dilation coefficient; The post-processing module is used to crop the output texture image to obtain a texture image that matches the size of the original fingerprint image.

7. A fingerprint texture extraction device, characterized in that: include: program; A memory, used for storing the program; A processor, used to load the program to execute the fingerprint texture extraction method based on deep learning as described in any one of claims 1 to 5.

8. A storage medium storing a program, characterized in that: When the program is executed by a processor, the fingerprint texture extraction method based on deep learning as described in any one of claims 1 to 5 is implemented.

Citation Information

Patent Citations

  • Fingerprint information processing method

    CN104809464A

  • Image segmentation method, system and electronic device based on depth learning

    CN109118491A

  • Semantic segmentation method based on efficient convolutional network and convolutional conditional random field

    CN110288603A

  • Texture image quality estimation method based on deep learning

    CN110400335A

  • Robust night image semantic segmentation method

    CN111008979A