CN114550224A - Fingerprint image identification comparison method and device based on deep learning and electronic equipment - Google Patents
Fingerprint image identification comparison method and device based on deep learning and electronic equipment Download PDFInfo
- Publication number
- CN114550224A CN114550224A CN202210051315.3A CN202210051315A CN114550224A CN 114550224 A CN114550224 A CN 114550224A CN 202210051315 A CN202210051315 A CN 202210051315A CN 114550224 A CN114550224 A CN 114550224A
- Authority
- CN
- China
- Prior art keywords
- fingerprint
- image
- fingerprint image
- neural network
- images
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
- Collating Specific Patterns (AREA)
Abstract
The invention provides a fingerprint image identification comparison method and device based on deep learning and an electronic device, wherein the method comprises the following steps: acquiring a first fingerprint image, a second fingerprint image and a plurality of fingerprint images to be compared; generating a fingerprint classification category according to the first fingerprint image; training a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model; training a second neural network according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model; and obtaining at least one image similarity according to the plurality of fingerprint images to be compared and the second network model, wherein the image similarity is used for representing the similarity of the two fingerprint images to be compared. The fingerprint classification classes are generated through the fingerprint images, the neural network models are trained by applying the fingerprint classification classes and different neural network structures, and the trained neural network models are applied to perform image identification comparison on the fingerprint images to be compared, so that the effect of improving the identification performance of the fingerprint images is realized.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a fingerprint image identification comparison method and device based on deep learning and electronic equipment.
Background
The fingerprint identification technology is widely applied at present, and can be used for collecting evidences from criminal sites by public security organs, authenticating the identity of bank clients, unlocking mobile phones or other intelligent equipment and the like. At present, a fingerprint identification algorithm is mainly based on a fingerprint feature point matching method.
In the prior art, a method based on feature point matching needs to accurately extract feature points and manually design a feature descriptor. Poor quality images have the problems of wrong or too few extracted feature points, and the quality of feature descriptors is also affected, so that the recognition performance is low. Therefore, improving the recognition performance of fingerprint images is an important problem to be solved urgently.
Disclosure of Invention
The invention provides a fingerprint image identification comparison method and device based on deep learning and electronic equipment, which are used for solving the problems that in the prior art, images with poor quality have errors or too few extracted feature points, the quality of feature descriptors is influenced, and the identification performance is low, and the effect of improving the identification performance of fingerprint images is realized.
The invention provides a fingerprint image identification comparison method based on deep learning, which comprises the following steps:
acquiring a first fingerprint image, a second fingerprint image and a plurality of fingerprint images to be compared;
generating a fingerprint classification category according to the first fingerprint image;
training a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model;
training a second neural network according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model;
and obtaining at least one image similarity according to the plurality of fingerprint images to be compared and the second network model, wherein the image similarity is used for representing the similarity of the two fingerprint images to be compared.
According to the fingerprint image identification comparison method based on deep learning, provided by the invention, a plurality of first fingerprint images are acquired from the same fingerprint;
the step of generating a fingerprint classification category from the first fingerprint image comprises:
carrying out image alignment on the plurality of first fingerprint images to obtain a first fingerprint image group;
and acquiring image blocks of the plurality of first fingerprint images at preset positions of the first fingerprint image group as fingerprint classification categories.
According to the fingerprint image identification comparison method based on deep learning provided by the invention, the step of aligning the plurality of first fingerprint images to obtain a first fingerprint image group comprises the following steps:
for each first fingerprint image, extracting fingerprint feature points by using a third neural network;
and aligning the fingerprint feature points of the first fingerprint images by using a standard three-dimensional data structure-based cylindrical coding algorithm to obtain a first fingerprint image group.
According to the fingerprint image identification comparison method based on deep learning, provided by the invention, a plurality of second fingerprint images are provided;
the step of training a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model comprises the following steps:
preprocessing the second fingerprint image to obtain a processed second fingerprint image;
inputting the processed second fingerprint image into the first neural network, and calculating a first cross entropy loss function, wherein the classification index of the first neural network is the fingerprint classification category;
when the first cross entropy loss function does not meet a first preset condition, optimizing parameters of the first neural network according to the first cross entropy loss function, returning to execute the second processed fingerprint image input to the first neural network again, and calculating the first cross entropy loss function;
and when the first cross entropy loss function meets the first preset condition, determining that the first neural network is a first network model.
According to the fingerprint image identification comparison method based on deep learning provided by the invention, the step of training a second neural network according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model comprises the following steps:
assigning weights of the full-connection classification layer of the first network model to the full-connection classification layer of the second neural network and freezing the full-connection classification layer of the second neural network;
inputting the processed second fingerprint image into the second neural network, and calculating a second cross entropy loss function, wherein the classification index of the second neural network is the fingerprint classification category;
when the second cross entropy loss function does not meet a second preset condition, optimizing parameters of the second neural network according to the second cross entropy loss function, returning to execute the second cross entropy loss function, inputting the processed second fingerprint image into the second neural network again, and calculating the second cross entropy loss function;
and when the second cross entropy loss function meets the second preset condition, determining the second neural network as a second network model.
According to the fingerprint image identification comparison method based on deep learning provided by the invention, the step of obtaining at least one image similarity according to the plurality of fingerprint images to be compared and the second network model comprises the following steps:
for each fingerprint image to be compared, sampling at intervals to obtain N image blocks to be predicted;
for each fingerprint image to be compared, inputting the N image blocks to be predicted into the second network model to obtain a feature vector set comprising N feature vectors;
calculating the vector pair similarity of N-by-N feature vector pairs in the two feature vector sets of the fingerprint images to be compared;
and calculating the image similarity of the two fingerprint images to be compared according to the vector pair similarity and Hungarian algorithm of the two fingerprint images to be compared.
According to the fingerprint image identification comparison method based on deep learning provided by the invention, after the step of calculating the image similarity of the two fingerprint images to be compared according to the vector pair similarity and Hungarian algorithm of the two fingerprint images to be compared, the method further comprises the following steps of:
and when the image similarity is greater than a preset threshold value, determining the two fingerprint images to be compared as associated fingerprint images.
The invention also provides a fingerprint image identification comparison device based on deep learning, which comprises:
the device comprises an acquisition unit, a comparison unit and a comparison unit, wherein the acquisition unit is used for acquiring a first fingerprint image, a second fingerprint image and a plurality of fingerprint images to be compared;
a category generating unit for generating a fingerprint classification category from the first fingerprint image;
the first training unit is used for training a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model;
the second training unit is used for training a second neural network according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model;
and the computing unit is used for obtaining at least one image similarity according to the plurality of fingerprint images to be compared and the second network model, and the image similarity and the similarity used for representing the two fingerprint images to be compared.
The invention also provides an electronic device, which comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein the processor executes the program to realize the steps of any one of the fingerprint image identification and comparison methods based on deep learning.
The present invention also provides a non-transitory computer readable storage medium, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the fingerprint image identification and comparison method based on deep learning as described in any one of the above.
The invention also provides a computer program product, which comprises a computer program, and the computer program is used for realizing the steps of the fingerprint image identification and comparison method based on deep learning.
According to the fingerprint image identification comparison method and device based on deep learning and the electronic equipment, a first fingerprint image, a second fingerprint image and a plurality of fingerprint images to be compared are obtained; generating a fingerprint classification category according to the first fingerprint image; training a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model; training a second neural network according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model; and obtaining a maximum similarity sum according to the plurality of fingerprint images to be compared and the second network model, wherein the maximum similarity sum is used for representing the similarity of different fingerprint images to be compared. The fingerprint classification classes are generated through the fingerprint images, the neural network models are trained by applying the fingerprint classification classes and different neural network structures, and the trained neural network models are applied to perform image identification comparison on the fingerprint images to be compared, so that the effect of improving the identification performance of the fingerprint images is realized.
Drawings
In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed for the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
FIG. 1 is a schematic flow chart of a deep learning-based fingerprint image identification comparison method according to the present invention;
FIG. 2 is a second schematic flowchart of a deep learning-based fingerprint image recognition and comparison method according to the present invention;
FIG. 3 is a third schematic flowchart of a deep learning-based fingerprint image recognition and comparison method according to the present invention;
FIG. 4 is a fourth schematic flowchart of a deep learning-based fingerprint image identification comparison method according to the present invention;
FIG. 5 is a schematic structural diagram of a deep learning-based fingerprint image recognition and comparison apparatus according to the present invention;
fig. 6 is a schematic structural diagram of an electronic device provided in the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The fingerprint identification technology is widely applied at present, and can be used for collecting evidences from criminal sites by public security organs, authenticating the identity of bank clients, unlocking mobile phones or other intelligent equipment and the like. At present, a fingerprint identification algorithm is mainly based on a fingerprint feature point matching method.
In the prior art, a method based on feature point matching needs to accurately extract feature points and manually design a feature descriptor. Poor quality images have the problems of wrong or too few extracted feature points and also influence the quality of feature descriptors, resulting in low recognition performance. Therefore, improving the recognition performance of fingerprint images is an important problem to be solved urgently.
In order to solve the above problem, the present invention provides a fingerprint image identification comparison method based on deep learning, as shown in fig. 1, including the following steps:
and S11, acquiring the first fingerprint image, the second fingerprint image and a plurality of fingerprint images to be compared.
And S12, generating a fingerprint classification category according to the first fingerprint image.
S13, training a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model.
S14, training a second neural network according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model.
S15, obtaining at least one image similarity according to the multiple fingerprint images to be compared and the second network model, wherein the image similarity is used for representing the similarity of the two fingerprint images to be compared.
In the embodiment of the invention, a first fingerprint image, a second fingerprint image and a plurality of fingerprint images to be compared are obtained, a fingerprint classification category is generated according to the first fingerprint image, a first neural network is trained according to the second fingerprint image and the fingerprint classification category to obtain a first network model, a second neural network is trained according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model, and at least one image similarity is obtained according to the plurality of fingerprint images to be compared and the second network model, wherein the image similarity is used for representing the similarity of the two fingerprint images to be compared. The fingerprint classification classes are generated through the fingerprint images, the neural network models are trained by applying the fingerprint classification classes and different neural network structures, and the trained neural network models are applied to perform image identification comparison on the fingerprint images to be compared, so that the effect of improving the identification performance of the fingerprint images is realized.
According to the fingerprint image identification comparison method based on deep learning, provided by the invention, a plurality of first fingerprint images are acquired from the same fingerprint; step S11 may include the steps of:
and S111, carrying out image alignment on the plurality of first fingerprint images to obtain a first fingerprint image group.
For convenience of understanding of the embodiments of the present invention, the following description will be made of the embodiments of the present invention with an electronic device as an execution subject, and the embodiments of the present invention are not limited thereto.
Specifically, the first fingerprint images are multiple, and the multiple first fingerprint images are collected from the same fingerprint. The first fingerprint image comprises a blank area and a fingerprint area, the electronic equipment can align the fingerprint areas in the first fingerprint images to obtain a group of aligned first fingerprint images as a first fingerprint image group.
And S112, acquiring image blocks of the plurality of first fingerprint images at preset positions of the first fingerprint image group as fingerprint classification categories.
Specifically, the electronic device may acquire a plurality of image blocks of the first fingerprint image at preset positions of the first fingerprint image group as fingerprint classification categories. The preset position can be set according to actual needs, and the preset position can be one or a plurality of.
In one example, the electronic device intercepts 80 × 80 small blocks at opposite alignment positions on different first fingerprint images of the same fingerprint, each of the first fingerprint images may intercept 2-6 small fingerprint blocks randomly, and the overlapping area of the different small fingerprint blocks is less than 5%. Each small fingerprint block serves as a new fingerprint classification category.
In the embodiment of the invention, the number of the first fingerprint images is multiple, the multiple first fingerprint images are acquired from the same fingerprint, the multiple first fingerprint images are subjected to image alignment to obtain a first fingerprint image group, and image blocks of the multiple first fingerprint images are acquired at the preset position of the first fingerprint image group to serve as fingerprint classification categories. The obtained fingerprint classification categories are derived from a plurality of different fingerprint images of the same fingerprint, so that the fingerprint classification categories are rich, and further, the first neural network and the second neural network are trained according to the fingerprint classification categories, so that the first network model and the second network model with higher accuracy can be obtained.
According to the fingerprint image identification comparison method based on deep learning provided by the invention, the step S111 may specifically include the following steps:
and S1111, for each first fingerprint image, extracting fingerprint feature points by using a third neural network.
Specifically, the electronic device may extract, for each first fingerprint image, fingerprint feature points in the first fingerprint image using a third neural network.
Preferably, the third neural network is a FingerNet network, and the electronic device can extract the fingerprint feature points in the first fingerprint image by using the FingerNet network.
S1112, aligning the fingerprint feature points of the first fingerprint images by using a standard three-dimensional data structure-based cylinder coding algorithm to obtain a first fingerprint image group.
Specifically, the electronic device may align the fingerprint feature points of a plurality of first fingerprint images by using a cylinder encoding algorithm based on a standard three-dimensional data structure, so as to obtain a first fingerprint image group.
In one example, for 10 first fingerprint images of one fingerprint collection, fingerprint feature points of the first fingerprint image are extracted from the 10 first fingerprint images by using a FingerNet network, the first fingerprint image with the most extracted fingerprint feature points is taken as a reference, fingerprint feature points of other first fingerprint images are aligned to the fingerprint feature points of the reference first fingerprint image by using a cylindrical coding algorithm based on a standard three-dimensional data structure, and then a group of aligned first fingerprint images is obtained and taken as a first fingerprint image group.
In the embodiment of the invention, for each first fingerprint image, a third neural network is used for extracting fingerprint feature points, and a standard three-dimensional data structure-based cylindrical coding algorithm is used for aligning the fingerprint feature points of a plurality of first fingerprint images to obtain a first fingerprint image group. The fingerprint classification method has the advantages that the fingerprint regions are arranged in the relative alignment positions of the first fingerprint images in the first fingerprint image group, no blank region exists, the fingerprint classification categories can be conveniently acquired subsequently according to the first fingerprint image group, the fingerprint classification categories are rich, the first neural network and the second neural network are trained according to the fingerprint classification categories, and the first network model and the second network model with higher accuracy can be obtained.
According to the fingerprint image identification comparison method based on deep learning, provided by the invention, a plurality of second fingerprint images are provided; as shown in fig. 2, step S13 may include the following specific steps:
and S21, preprocessing the second fingerprint image to obtain a processed second fingerprint image.
Specifically, the electronic device may perform preprocessing such as one or more of grayscale conversion, cropping, and rotation on the plurality of second fingerprint images to obtain a processed second fingerprint image.
S22, inputting the processed second fingerprint image into the first neural network, and calculating a first cross entropy loss function, wherein the classification index of the first neural network is the fingerprint classification category.
Specifically, the electronic device may input the processed second fingerprint image into a first neural network, and calculate a first cross entropy loss function, where a classification index of the first neural network is a fingerprint classification category.
Preferably, the first neural network is a relatively complex Resnest101 network, and the electronic device may input the processed second fingerprint image into the Resnest101 network to calculate the first cross entropy loss function.
In an example, in the training process, the renest 101 network may randomly intercept an image block with a size of 64 × 64 from the processed second fingerprint image, randomly rotate the image block by 30 degrees left and right to obtain a new image block, and randomly perform gray value stretching transformation on the image block to obtain the new image block. And finally, maximizing the classification boundary on an angle space by simultaneously normalizing the weight and the features of the fully connected classification layer to obtain a plurality of 128-dimensional feature vectors and calculating a first cross entropy loss function, wherein the plurality of 128-dimensional feature vectors are used for representing the classification result of the processed second fingerprint image on the fingerprint classification category through a Resnest101 network.
S23, when the first cross entropy loss function does not meet a first preset condition, optimizing the parameters of the first neural network according to the first cross entropy loss function, and returning to execute the step S22 again.
Specifically, when the first cross entropy loss function does not satisfy the first preset condition, the electronic device may optimize parameters of the first neural network according to the first cross entropy loss function, and return to re-execute the step of inputting the processed second fingerprint image into the first neural network and calculating the first cross entropy loss function.
The first preset condition may be that the first cross entropy loss function reaches a preset first cross entropy loss function threshold, and the first preset condition may also be that the falling speed of the first cross entropy loss function reaches a preset first falling speed threshold.
S24, when the first cross entropy loss function meets the first preset condition, determining that the first neural network is a first network model.
Specifically, the electronic device may determine that the first neural network is the first network model when the first cross entropy loss function satisfies a first preset condition.
In the embodiment of the invention, the second fingerprint image after processing is obtained by preprocessing the second fingerprint image, the second fingerprint image after processing is input into a first neural network, a first cross entropy loss function is calculated, the classification index of the first neural network is a fingerprint classification category, when the first cross entropy loss function does not meet a first preset condition, the parameters of the first neural network are optimized according to the first cross entropy loss function, the second fingerprint image after processing is input into the first neural network again, the first cross entropy loss function is calculated, and when the first cross entropy loss function meets the first preset condition, the first neural network is determined to be a first network model. The first network model is enabled to have higher accuracy in classification performance.
According to the fingerprint image identification comparison method based on deep learning provided by the present invention, as shown in fig. 3, step S14 may include the following steps:
s31, assigning the weight of the full-connection classification layer of the first network model to the full-connection classification layer of the second neural network and freezing the full-connection classification layer of the second neural network.
Specifically, the electronic device may assign a weight of the full-connection classification layer of the first network model to the full-connection classification layer of the second neural network, and freeze the full-connection classification layer of the second neural network, so that the full-connection classification layer of the second neural network is not updated during subsequent training of the second neural network.
Preferably, the second neural network is a less complex MobileNetV3 network, and the electronic device can assign weights for the fully connected classification layer of the first network model to the MobileNetV3 network and freeze the fully connected classification layer of the MobileNetV3 network.
S32, inputting the processed second fingerprint image into the second neural network, and calculating a second cross entropy loss function, wherein the classification index of the second neural network is the fingerprint classification category.
Specifically, the electronic device may input the processed second fingerprint image into a second neural network, and calculate a second cross entropy loss function, where a classification index of the second neural network is a fingerprint classification category.
Preferably, the electronic device can input the processed second fingerprint image into a MobileNetV3 network, and calculate a second cross entropy loss function.
In an example, in the training process, the MobileNetV3 network may randomly intercept an image block with a size of 64 × 64 from the processed second fingerprint image, randomly rotate the image block by 30 degrees left and right to obtain a new image block, and randomly perform gray value stretching transformation on the image block to obtain the new image block. And finally, maximizing a classification boundary on an angle space by simultaneously normalizing the weight and the features of the fully connected classification layer to obtain a plurality of 128-dimensional feature vectors and calculating a second cross entropy loss function, wherein the plurality of 128-dimensional feature vectors are used for representing the classification result of the processed second fingerprint image on the fingerprint classification category through a MobileNet V3 network.
S33, when the second cross entropy loss function does not meet a second preset condition, optimizing the parameters of the second neural network according to the second cross entropy loss function, and returning to execute the step S32 again.
Specifically, when the second cross entropy loss function does not satisfy the second preset condition, the electronic device may optimize parameters of the second neural network according to the second cross entropy loss function, and return to re-execute the step of inputting the processed second fingerprint image into the second neural network and calculating the second cross entropy loss function.
The second preset condition may be that the second cross entropy loss function reaches a preset second cross entropy loss function threshold, and the second preset condition may also be that the falling speed of the second cross entropy loss function reaches a preset second falling speed threshold.
S34, when the second cross entropy loss function meets the second preset condition, determining that the second neural network is a second network model.
Specifically, the electronic device may determine the second neural network as the second network model when the second cross entropy loss function satisfies the second preset condition.
In the embodiment of the invention, the weight of the full-connection classification layer of the first network model is assigned to the full-connection classification layer of the second neural network, the full-connection classification layer of the second neural network is frozen, the processed second fingerprint image is input into the second neural network, a second cross entropy loss function is calculated, the classification index of the second neural network is the fingerprint classification category, when the second cross entropy loss function does not meet a second preset condition, the parameters of the second neural network are optimized according to the second cross entropy loss function, the second neural network is input into the second neural network after the second fingerprint image is returned to be executed again, the second cross entropy loss function is calculated, and when the second cross entropy loss function meets the second preset condition, the second neural network is determined to be the second network model. And assigning the fully-connected classification layer of the first network model to a fully-connected classification layer of a second neural network with relatively low complexity, aligning the features of the second neural network and the first neural network, and training on the basis to obtain a second network model, so that the second network model with relatively low complexity has higher accuracy in classification performance, and has the effect of higher recognition speed compared with the first network model.
According to the fingerprint image identification comparison method based on deep learning provided by the present invention, as shown in fig. 4, step S15 may include the following steps:
and S41, for each fingerprint image to be compared, sampling at intervals to obtain N image blocks to be predicted.
Specifically, the electronic device may perform interval sampling on each fingerprint image to be compared, and obtain N image blocks to be predicted from one fingerprint image to be compared.
In one example, the electronic device samples the fingerprint image to be compared at intervals, the interval pixel value of each sample point is 16, and a small image block with a size of 64 × 64 pixel values is taken as the image block to be predicted by taking each sample point as a center.
And S42, inputting the N image blocks to be predicted into the second network model for each fingerprint image to be compared to obtain a feature vector set comprising N feature vectors.
Specifically, for each fingerprint image to be compared, the electronic device may input the N image blocks to be predicted into the second network model to obtain a feature vector set including N feature vectors.
In one example, the electronic device inputs to-be-predicted image blocks of a to-be-compared fingerprint image into a second network model MobileNetV3 network, each to-be-predicted image block is classified by a MobileNetV3 network to generate a 128-dimensional feature vector, and inputs N to-be-predicted image blocks of a to-be-compared fingerprint image into a second network model MobileNetV3 network to obtain a feature vector set including N feature vectors.
And S43, calculating the vector pair similarity of N x N feature vector pairs for the two feature vector sets of the two fingerprint images to be compared.
Specifically, two fingerprint images to be compared correspond to two feature sets. The electronic device may calculate the vector pair similarity of N × N feature vector pairs according to the two feature vector sets of the two fingerprint images to be compared.
In one example, the electronic device calculates the similarity of each pair of all features in the two feature vector sets to obtain the vector pair similarity of N × N feature vector pairs, and the vector pair similarity may be calculated by using a feature dot product.
S44, calculating the image similarity of the two fingerprint images to be compared according to the vector pair similarity of the two fingerprint images to be compared and the Hungarian algorithm.
Specifically, the electronic device can calculate the image similarity of the two fingerprint images to be compared according to the vector pair similarity and the hungarian algorithm of the two fingerprint images to be compared.
In one example, the electronic device may sort N × N vector pair similarities of the two fingerprint images to be compared from large to small, take a feature vector pair corresponding to a preset number of vector pair similarities in a sequence, obtain an alignment matrix by using a hungarian algorithm, and calculate the image similarity of the two fingerprint images to be compared according to the alignment matrix.
In the embodiment of the invention, N image blocks to be predicted are obtained by sampling every fingerprint image to be compared at intervals, the N image blocks to be predicted are input into the second network model for every fingerprint image to be compared to obtain a feature vector set containing N feature vectors, the vector pair similarity of N characteristic vector pairs in the two feature vector sets of the two fingerprint images to be compared is calculated, and the image similarity of the two fingerprint images to be compared is calculated according to the vector pair similarity of the two fingerprint images to be compared and the Hungarian algorithm. The vector pair similarity is calculated through the N feature vector sets of the fingerprint images to be compared, and the image similarity is calculated according to the vector pair similarity, so that the image similarity of the two fingerprint images to be compared can be accurately and efficiently calculated.
According to the fingerprint image identification comparison method based on deep learning provided by the invention, after the step S15, the method further comprises:
and when the image similarity is greater than a preset threshold value, determining the two fingerprint images to be compared as associated fingerprint images.
Specifically, when the image similarity of the two fingerprint images to be compared is greater than a preset threshold, the electronic device may determine that the two fingerprint images to be compared are associated fingerprint images. The preset threshold value can be set according to actual needs.
In the embodiment of the invention, after the image similarity of the two fingerprint images to be compared is determined, when the image similarity is greater than a preset threshold value, the two fingerprint images to be compared are determined to be associated fingerprint images. The effect of efficiently determining the associated fingerprint image is achieved.
The fingerprint image identification and comparison device based on deep learning provided by the invention is described below, and the fingerprint image identification and comparison device based on deep learning described below and the fingerprint image identification and comparison method based on deep learning described above can be referred to correspondingly.
The invention also provides a fingerprint image identification comparison device based on deep learning, as shown in fig. 5, comprising:
the acquiring unit 51 is configured to acquire a first fingerprint image, a second fingerprint image, and a plurality of fingerprint images to be compared.
A class generating unit 52 for generating a fingerprint classification class from the first fingerprint image.
And the first training unit 53 is configured to train a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model.
And a second training unit 54, configured to train a second neural network according to the second fingerprint image, the fingerprint classification category, and the first network model, so as to obtain a second network model.
And the calculating unit 55 is configured to obtain at least one image similarity according to the plurality of fingerprint images to be compared and the second network model, where the image similarity is used to represent the similarity between two fingerprint images to be compared.
In the embodiment of the invention, a first fingerprint image, a second fingerprint image and a plurality of fingerprint images to be compared are obtained, a fingerprint classification category is generated according to the first fingerprint image, a first neural network is trained according to the second fingerprint image and the fingerprint classification category to obtain a first network model, a second neural network is trained according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model, and at least one image similarity is obtained according to the plurality of fingerprint images to be compared and the second network model, wherein the image similarity is used for representing the similarity of the two fingerprint images to be compared. The fingerprint classification classes are generated through the fingerprint images, the neural network model is trained by the fingerprint classification classes and different neural network structures, and the trained neural network model is applied to carry out image identification comparison on the fingerprint images to be compared, so that the effect of improving the identification performance of the fingerprint images is realized.
According to the fingerprint image identification comparison device based on deep learning, provided by the invention, a plurality of first fingerprint images are acquired from the same fingerprint;
a category generating unit 52, configured to perform image alignment on the multiple first fingerprint images to obtain a first fingerprint image group; and acquiring image blocks of the plurality of first fingerprint images at preset positions of the first fingerprint image group as fingerprint classification categories.
According to the fingerprint image identification comparison device based on deep learning provided by the invention, the category generating unit 52 is specifically configured to extract a fingerprint feature point for each first fingerprint image by using a third neural network; and aligning the fingerprint feature points of the first fingerprint images by using a standard three-dimensional data structure-based cylindrical coding algorithm to obtain a first fingerprint image group.
According to the fingerprint image identification comparison device based on deep learning, provided by the invention, a plurality of second fingerprint images are provided;
the first training unit 53 is specifically configured to perform preprocessing on the second fingerprint image to obtain a processed second fingerprint image; inputting the processed second fingerprint image into the first neural network, and calculating a first cross entropy loss function, wherein the classification index of the first neural network is the fingerprint classification category; when the first cross entropy loss function does not meet a first preset condition, optimizing parameters of the first neural network according to the first cross entropy loss function, returning to execute the second processed fingerprint image input to the first neural network again, and calculating the first cross entropy loss function; and when the first cross entropy loss function meets the first preset condition, determining that the first neural network is a first network model.
According to the fingerprint image recognition and comparison device based on deep learning provided by the invention, the second training unit 54 is specifically used for assigning the weight of the full-connection classification layer of the first network model to the full-connection classification layer of the second neural network and freezing the full-connection classification layer of the second neural network; inputting the processed second fingerprint image into the second neural network, and calculating a second cross entropy loss function, wherein the classification index of the second neural network is the fingerprint classification category; when the second cross entropy loss function does not meet a second preset condition, optimizing parameters of the second neural network according to the second cross entropy loss function, returning to execute the second cross entropy loss function, inputting the processed second fingerprint image into the second neural network again, and calculating the second cross entropy loss function; and when the second cross entropy loss function meets the second preset condition, determining the second neural network as a second network model.
According to the fingerprint image identification comparison device based on deep learning provided by the invention, the calculation unit 55 is specifically configured to sample each fingerprint image to be compared at intervals to obtain N image blocks to be predicted; for each fingerprint image to be compared, inputting the N image blocks to be predicted into the second network model to obtain a feature vector set comprising N feature vectors; calculating the vector pair similarity of N-by-N feature vector pairs in two feature vector sets of the two fingerprint images to be compared; and calculating the image similarity of the two fingerprint images to be compared according to the vector pair similarity of the two fingerprint images to be compared and the Hungarian algorithm.
According to the fingerprint image identification comparison device based on deep learning provided by the invention, the calculation unit 55 is further configured to determine that the two fingerprint images to be compared are related fingerprint images when the image similarity is greater than a preset threshold.
Fig. 6 illustrates a physical structure diagram of an electronic device, which may include, as shown in fig. 6: a processor (processor)610, a communication Interface (Communications Interface)620, a memory (memory)630 and a communication bus 640, wherein the processor 610, the communication Interface 620 and the memory 630 communicate with each other via the communication bus 640. The processor 610 may invoke logic instructions in the memory 630 to perform a deep learning based fingerprint image identification comparison method comprising: acquiring a first fingerprint image, a second fingerprint image and a plurality of fingerprint images to be compared; generating a fingerprint classification category according to the first fingerprint image; training a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model; training a second neural network according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model; and obtaining at least one image similarity according to the plurality of fingerprint images to be compared and the second network model, wherein the image similarity is used for representing the similarity of the two fingerprint images to be compared.
In addition, the logic instructions in the memory 630 may be implemented in software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In another aspect, the present invention further provides a computer program product, where the computer program product includes a computer program, the computer program may be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, a computer can execute the deep learning based fingerprint image identification comparison method provided by the above methods, where the method includes: acquiring a first fingerprint image, a second fingerprint image and a plurality of fingerprint images to be compared; generating a fingerprint classification category according to the first fingerprint image; training a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model; training a second neural network according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model; and obtaining at least one image similarity according to the plurality of fingerprint images to be compared and the second network model, wherein the image similarity is used for representing the similarity of the two fingerprint images to be compared.
In yet another aspect, the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, the computer program being implemented by a processor to execute the deep learning-based fingerprint image identification comparison method provided by the above methods, the method including: acquiring a first fingerprint image, a second fingerprint image and a plurality of fingerprint images to be compared; generating a fingerprint classification category according to the first fingerprint image; training a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model; training a second neural network according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model; and obtaining at least one image similarity according to the plurality of fingerprint images to be compared and the second network model, wherein the image similarity is used for representing the similarity of the two fingerprint images to be compared. .
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (11)
1. A fingerprint image identification comparison method based on deep learning is characterized by comprising the following steps:
acquiring a first fingerprint image, a second fingerprint image and a plurality of fingerprint images to be compared;
generating a fingerprint classification category according to the first fingerprint image;
training a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model;
training a second neural network according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model;
and obtaining at least one image similarity according to the plurality of fingerprint images to be compared and the second network model, wherein the image similarity is used for representing the similarity of the two fingerprint images to be compared.
2. The deep learning based fingerprint image identification comparison method according to claim 1, wherein the first fingerprint images are multiple, and the multiple first fingerprint images are collected from the same fingerprint;
the step of generating a fingerprint classification category from the first fingerprint image comprises:
carrying out image alignment on the plurality of first fingerprint images to obtain a first fingerprint image group;
and acquiring image blocks of the plurality of first fingerprint images at preset positions of the first fingerprint image group as fingerprint classification categories.
3. The deep learning-based fingerprint image identification comparison method according to claim 2, wherein the step of performing image alignment on the plurality of first fingerprint images to obtain a first fingerprint image group comprises:
for each first fingerprint image, extracting fingerprint feature points by using a third neural network;
and aligning the fingerprint feature points of the first fingerprint images by using a standard three-dimensional data structure-based cylindrical coding algorithm to obtain a first fingerprint image group.
4. The deep learning based fingerprint image identification comparison method according to claim 1, wherein the second fingerprint image is a plurality of fingerprint images;
the step of training a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model comprises the following steps:
preprocessing the second fingerprint image to obtain a processed second fingerprint image;
inputting the processed second fingerprint image into the first neural network, and calculating a first cross entropy loss function, wherein the classification index of the first neural network is the fingerprint classification category;
when the first cross entropy loss function does not meet a first preset condition, optimizing parameters of the first neural network according to the first cross entropy loss function, returning to execute the second processed fingerprint image input to the first neural network again, and calculating the first cross entropy loss function;
and when the first cross entropy loss function meets the first preset condition, determining that the first neural network is a first network model.
5. The deep learning based fingerprint image recognition and comparison method according to claim 4, wherein the step of training a second neural network according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model comprises:
assigning weights of the full-connection classification layer of the first network model to the full-connection classification layer of the second neural network and freezing the full-connection classification layer of the second neural network;
inputting the processed second fingerprint image into the second neural network, and calculating a second cross entropy loss function, wherein the classification index of the second neural network is the fingerprint classification category;
when the second cross entropy loss function does not meet a second preset condition, optimizing parameters of the second neural network according to the second cross entropy loss function, returning to execute the second cross entropy loss function, inputting the processed second fingerprint image into the second neural network again, and calculating the second cross entropy loss function;
and when the second cross entropy loss function meets the second preset condition, determining the second neural network as a second network model.
6. The deep learning-based fingerprint image identification comparison method according to claim 1, wherein the step of obtaining at least one image similarity according to the plurality of fingerprint images to be compared and the second network model comprises:
for each fingerprint image to be compared, sampling at intervals to obtain N image blocks to be predicted;
for each fingerprint image to be compared, inputting the N image blocks to be predicted into the second network model to obtain a feature vector set comprising N feature vectors;
calculating the vector pair similarity of N-by-N feature vector pairs in two feature vector sets of the two fingerprint images to be compared;
and calculating the image similarity of the two fingerprint images to be compared according to the vector pair similarity of the two fingerprint images to be compared and the Hungarian algorithm.
7. The deep learning based fingerprint image identification comparison method as claimed in claim 6, wherein after the step of calculating the image similarity of the two fingerprint images to be compared according to the vector pair similarity and Hungarian algorithm of the two fingerprint images to be compared, further comprising:
and when the image similarity is greater than a preset threshold value, determining the two fingerprint images to be compared as associated fingerprint images.
8. A fingerprint image identification comparison device based on deep learning is characterized by comprising:
the device comprises an acquisition unit, a comparison unit and a comparison unit, wherein the acquisition unit is used for acquiring a first fingerprint image, a second fingerprint image and a plurality of fingerprint images to be compared;
a category generating unit for generating a fingerprint classification category from the first fingerprint image;
the first training unit is used for training a first neural network according to the second fingerprint image and the fingerprint classification category to obtain a first network model;
the second training unit is used for training a second neural network according to the second fingerprint image, the fingerprint classification category and the first network model to obtain a second network model;
and the calculating unit is used for obtaining at least one image similarity according to the plurality of fingerprint images to be compared and the second network model, and the image similarity is used for representing the similarity of the two fingerprint images to be compared.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the steps of the deep learning based fingerprint image recognition and comparison method according to any one of claims 1 to 7.
10. A non-transitory computer readable storage medium, on which a computer program is stored, wherein the computer program, when being executed by a processor, implements the steps of the deep learning based fingerprint image recognition comparison method according to any one of claims 1 to 7.
11. A computer program product comprising a computer program, wherein the computer program is configured to implement the steps of the method for fingerprint image recognition and comparison based on deep learning according to any one of claims 1 to 7 when executed by a processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210051315.3A CN114550224A (en) | 2022-01-17 | 2022-01-17 | Fingerprint image identification comparison method and device based on deep learning and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210051315.3A CN114550224A (en) | 2022-01-17 | 2022-01-17 | Fingerprint image identification comparison method and device based on deep learning and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114550224A true CN114550224A (en) | 2022-05-27 |
Family
ID=81670699
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210051315.3A Pending CN114550224A (en) | 2022-01-17 | 2022-01-17 | Fingerprint image identification comparison method and device based on deep learning and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114550224A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116386091A (en) * | 2022-11-18 | 2023-07-04 | 荣耀终端有限公司 | Fingerprint identification method and device |
-
2022
- 2022-01-17 CN CN202210051315.3A patent/CN114550224A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116386091A (en) * | 2022-11-18 | 2023-07-04 | 荣耀终端有限公司 | Fingerprint identification method and device |
| CN116386091B (en) * | 2022-11-18 | 2024-04-02 | 荣耀终端有限公司 | Fingerprint identification method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yuan et al. | Fingerprint liveness detection using an improved CNN with image scale equalization | |
| US9514356B2 (en) | Method and apparatus for generating facial feature verification model | |
| CN104778457B (en) | Video face identification method based on multi-instance learning | |
| CN106951832B (en) | Verification method and device based on handwritten character recognition | |
| CN109190470B (en) | Pedestrian re-identification method and device | |
| CN112801054B (en) | Face recognition model processing method, face recognition method and device | |
| CN108564040B (en) | Fingerprint activity detection method based on deep convolution characteristics | |
| Mistry et al. | Fingerprint synthesis: Search with 100 million prints | |
| Jena et al. | Improved offline signature verification scheme using feature point extraction method | |
| CN111985332B (en) | Gait recognition method of improved loss function based on deep learning | |
| CN113095156A (en) | Double-current network signature identification method and device based on inverse gray scale mode | |
| CN110414431B (en) | Face recognition method and system based on elastic context relation loss function | |
| CN114550224A (en) | Fingerprint image identification comparison method and device based on deep learning and electronic equipment | |
| Omri et al. | Cloud-based mobile system for biometrics authentication | |
| CN111488798A (en) | Fingerprint identification method and device, electronic equipment and storage medium | |
| WO2021189364A1 (en) | Method and device for generating adversarial image, equipment, and readable storage medium | |
| Kadhm et al. | An accurate signature verification system based on proposed HSC approach and ANN architecture | |
| CN112733670A (en) | Fingerprint feature extraction method and device, electronic equipment and storage medium | |
| CN111428643A (en) | Finger vein image recognition method and device, computer equipment and storage medium | |
| Schmidt et al. | Dynamic signature recognition based on fisher discriminant | |
| CN113536870A (en) | Abnormal head portrait identification method and device | |
| Choubey et al. | Bilateral Partitioning based character recognition for Vehicle License plate | |
| Chirchi et al. | Modified circular fuzzy segmentor and local circular encoder to iris segmentation and recognition | |
| CN112926041B (en) | Remote identity authentication system based on biological characteristics | |
| CN117118765B (en) | IPV6 identity security authentication method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |