An efficient fingerprint recognition system and method based on Faiss library
Through the efficient fingerprint recognition system based on the Faiss library, using vector compression and matrix conversion technology, the problem of inefficient retrieval efficiency in AFIS caused by different number of fine nodes is solved, efficient fingerprint matching and comparison is achieved, and the system is improved on a large scale Processing capabilities in the database.
Patent Information
- Application Number
- CN202111492301.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-08
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2041-12-08
AI Technical Summary
When facing a huge database, how to improve matching efficiency while ensuring accuracy is an urgent problem, especially the retrieval efficiency caused by the different number of fine nodes in each fingerprint. low.
Using an efficient fingerprint recognition system based on the Faiss library, the fingerprint image to be identified is converted into a fixed-length fingerprint representation vector through the feature extraction module, and the index type provided by the Faiss search library is accelerated, combining vector compression and matrix conversion technology to convert the original Detailed node encoding is dense and binarized, maximizing the correlation between preservation of representation bits and achieving efficient fingerprint matching.
It significantly improves the efficiency of fingerprint comparison, especially during peak transactions, which can significantly improve the system's processing capabilities and improve the efficiency of the entire transaction process.
Smart Images

Figure CN113989867B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fingerprint identification, and in particular to an efficient fingerprint identification system and method based on a Faiss library. Background Art
[0002] Fingerprints are a biometric feature that is easy to obtain, unique, and lifelong, providing a reliable way to identify people. The modern fingerprint recognition technology originated from the research of Galton and others. Automatic fingerprint identification systems (AFIS) are widely used in forensic and government applications, but with the growth of the population, the size of modern AFIS databases has exceeded billions. How to improve matching efficiency while ensuring accuracy is an urgent problem to be solved.
[0003] Minutiae points are the main features used for fingerprint matching. Currently, the most applicable and effective fingerprint recognition technology recognized is based on MCC coding, supplemented by indexing technology to achieve 1:N matching. However, the problem is that the number of minutiae points in each fingerprint is different. Even for the fingerprints of the same finger, the number of minutiae points extracted may vary due to reasons such as the collection method or the finger state. At present, there are two types of improvements based on this. One is to compress the dimension represented by each minutiae point, and the other is to represent the characteristics of each finger with a fixed length, which is beneficial to improve the retrieval speed. Summary of the invention
[0004] In view of the problems existing in the prior art, the present invention provides an efficient fingerprint recognition system and method based on the Faiss library.
[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:
[0006] On the one hand, an efficient fingerprint recognition system based on Faiss library includes a basic service module, a feature value management module, and a fingerprint retrieval module.
[0007] The basic service module accepts a fingerprint matching request in the form of a fingerprint collection image to be identified and a fingerprint library ID to be retrieved, and sends the fingerprint image to be identified to the feature value management module;
[0008] The feature value management module includes a feature extraction module and a feature value storage module, wherein the feature extraction module receives the fingerprint image to be identified, extracts the feature value of the fingerprint image, obtains a fixed-length fingerprint representation vector, and the feature value storage module stores the generated feature value;
[0009] The fingerprint retrieval module provides an index type to speed up retrieval.
[0010] On the other hand, an efficient fingerprint recognition method based on the Faiss library is implemented by the aforementioned efficient fingerprint recognition system based on the Faiss library, and specifically comprises the following steps:
[0011] Step 1: Use a fingerprint collection instrument to collect sample fingerprint images, send the sample fingerprint images to the feature extraction module for feature extraction, obtain sample fingerprint feature values, and store them in the feature value storage module to create a sample fingerprint library;
[0012] The fingerprint database stores the fingerprint image, fingerprint owner ID, fingerprint ID, and fingerprint representation as an entry in the database;
[0013] Step 2: The basic service module accepts the fingerprint matching request from the client and sends the fingerprint image to be identified to the feature extraction module;
[0014] Step 3: The feature extraction module extracts the minutiae of the fingerprint image to be identified and converts the minutiae into MCC codes;
[0015] Step 4: The fingerprint retrieval module interacts with the feature value storage module to perform vector compression on the MCC code to make the code dense, and compares the feature value obtained in step 3 with the sample fingerprint feature value in the fingerprint library to obtain the ID corresponding to the fingerprint feature value with the highest similarity and its similarity score;
[0016] The fingerprint retrieval module sets a threshold for the similarity score. If the highest score exceeds the threshold, it is considered that the retrieval fails and there is no matching fingerprint image in the fingerprint database.
[0017] The vector compression uses matrix transformation to reduce the dimension of the original detail point code and binarize it at the same time, wherein the matrix used for transformation is obtained through learning, and a Laplacian matrix is used to maximize the correlation between the representation bits. Finally, each fingerprint detail point is represented as a 128-bit binary representation, and each fingerprint image corresponds to a binary vector set;
[0018] Step 5: Return the fingerprint comparison result to the client that initiated the search request. If the search is unsuccessful, the return value is empty, thus realizing fingerprint recognition.
[0019] The fingerprint comparison result is the fingerprint image with the highest similarity in step 4 and its corresponding ID;
[0020] The beneficial effects produced by the present invention are:
[0021] The present invention proposes an efficient fingerprint recognition system and method based on the Faiss library. The present invention is applied to fingerprint recognition tasks, especially when high requirements are placed on the comparison efficiency, so that efficient retrieval can be performed to improve the efficiency of the entire transaction process. When facing the peak period of transactions, the improvement of transaction efficiency is urgently needed in the environment. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 It is a structural diagram of the fingerprint recognition system of the present invention.
[0023] Figure 2 This is a flow chart of the fingerprint recognition method of the present invention. DETAILED DESCRIPTION
[0024] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.
[0025] On the one hand, an efficient fingerprint recognition system based on the Faiss library, such as Figure 1 As shown, it includes a basic service module, a feature value management module, and a fingerprint retrieval module.
[0026] The basic service module accepts a fingerprint matching request in the form of a fingerprint collection image to be identified and a fingerprint library ID to be retrieved, and sends the fingerprint image to be identified to the feature value management module;
[0027] The eigenvalue management module includes a feature extraction module and a eigenvalue storage module, wherein the feature extraction module receives the fingerprint image to be identified, extracts the fingerprint image eigenvalue, obtains the fixed-length fingerprint representation vector, and first extracts the MCC detail point representation. Since the representation vector of each MCC detail point is of fixed length, the number of MCC detail points contained in the fingerprint images belonging to the same fingerprint is not exactly the same. To this end, the 1280-bit MCC representation is first compressed, and then, through spectral clustering, for each detail point, 1 is added to the bit according to which category it belongs to. The eigenvalue storage module stores the generated eigenvalues;
[0028] The fingerprint retrieval module is built based on the open source Faiss of Facebook. The Faiss retrieval library provides index types to speed up retrieval. Here, LSH is called in this embodiment.
[0029] On the other hand, an efficient fingerprint recognition method based on the Faiss library, such as Figure 2 As shown, the above-mentioned efficient fingerprint recognition system based on the Faiss library is implemented, which specifically includes the following steps:
[0030] Step 1: Use a fingerprint collection instrument to collect sample fingerprint images and create a fingerprint library;
[0031] When adding the sample fingerprint image of fingerprint collection, the fingerprint library sends the sample fingerprint image to the feature extraction module for feature extraction, obtains the sample fingerprint feature value, and stores it in the feature value storage module; wherein one finger corresponds to several collection images.
[0032] The fingerprint database stores fingerprint images, the ID of the fingerprint owner, the fingerprint ID, and the fingerprint representation as an entry in the database. One finger can correspond to multiple fingerprint images, and the corresponding fingerprint IDs are the same.
[0033] Step 2: The basic service module receives the fingerprint matching request sent by the client and sends the fingerprint image to be recognized to the feature extraction module.
[0034] Step 3: The feature extraction module extracts the minutiae of the fingerprint image to be recognized and converts the minutiae into MCC codes.
[0035] When each fingerprint image to be recognized is input, the feature extraction module extracts minutiae features (ridge lines and bifurcation points), and then sends them to the feature encoding module to generate the original MCC codes. The vector compression module is further executed to densify the original MCC codes through matrix transformation, while maximizing the retention of the correlation between minutiae and expanding the space utilization rate. The processed fingerprint image representation is stored in the fingerprint database and identified by the ID of the fingerprint owner.
[0036] Step 4: The fingerprint retrieval module interacts with the eigenvalue storage module to vector compress the MCC codes to make the codes denser, compares the eigenvalues obtained in Step 3 with the sample fingerprint eigenvalues in the fingerprint database, and obtains the ID corresponding to the fingerprint eigenvalue with the highest similarity and its similarity score.
[0037] The fingerprint retrieval module sets a threshold for the similarity score. If the highest score exceeds the threshold, it is considered that the retrieval fails and there is no fingerprint image in the fingerprint database that can be matched.
[0038] The vector compression uses matrix transformation to reduce the dimension of the original minutiae encoding and binarize it at the same time. Among them, the matrix used for transformation is obtained through learning, and a Laplacian matrix is used to maximize the retention of the correlation between representation bits. Finally, each fingerprint minutia is represented as a 128-bit binary representation, and each fingerprint image corresponds to a set of binary vectors.
[0039] Specifically, first, the original eigenvalue matrix X ∈ R d×n , representing n minutiae contained in a fingerprint image, and each minutia is represented by d dimensions. By performing matrix transformation with the matrix W ∈ R l×d , l < n, the matrix D ∈ R l ×n is obtained, which is the set of compressed eigenvectors.
[0040] Among them, the learning algorithm of the matrix W is as follows:
[0041] Initialize W: Use XX TThe first l eigenvectors of construct the matrix W;
[0042] Update W: Used to binarize the matrix generated after conversion;
[0043] Learning objective function:
[0044] Fixed B, L is the Laplace matrix.
[0045] Specifically, the target matrix W is learned by the gradient descent algorithm.
[0046] Specifically, the similarity score is calculated as:
[0047] Similarity Where q is the code length, which is 128 here;
[0048] Score f score (V1,V2)=Σmaxf s (v,v j ).
[0049] Step 5: Return the fingerprint comparison result to the client that initiated the search request. If the search is unsuccessful, the return value is empty.
[0050] The fingerprint comparison result is the fingerprint image with the highest similarity in step 4 and its corresponding ID;
[0051] The fingerprint retrieval module, i.e., the similarity search tool based on Faiss, is applied to the compressed detail point set. Faiss mainly accelerates indexing by compressing the input vector. When the present invention performs the compression process one more time in advance, the retrieval speed will obviously be further improved. This module uses Faiss to provide an interface, and is connected to the created fingerprint library. The Faiss similarity search tool is applied to the fingerprint feature value vector group to be compared with the vector group of vectors saved in the fingerprint library. Finally, TOPk is selected from the fingerprint image with the highest similarity score as the input of the next module.
[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.
Claims
1. An efficient fingerprint recognition system based on Faiss library, characterized in that: It includes basic service module, feature value management module and fingerprint retrieval module; The basic service module accepts a fingerprint matching request in the form of a fingerprint collection image to be identified and a fingerprint library ID to be retrieved, and sends the fingerprint image to be identified to the feature value management module; The feature value management module extracts the feature value of the fingerprint image, stores the feature value and compares it with the feature value of the sample fingerprint; The fingerprint retrieval module provides an index type to speed up retrieval; The feature value management module includes a feature extraction module and a feature value storage module, wherein the feature extraction module receives the fingerprint image to be identified, extracts the feature value of the fingerprint image, obtains a fixed-length fingerprint representation vector, and the feature value storage module stores the generated feature value; The efficient fingerprint recognition system based on the Faiss library is implemented to realize an efficient fingerprint recognition method based on the Faiss library, comprising the following steps: Step 1: Use a fingerprint collection instrument to collect sample fingerprint images, send the sample fingerprint images to the feature extraction module for feature extraction, obtain sample fingerprint feature values, and store them in the feature value storage module to create a sample fingerprint library; The fingerprint database stores the fingerprint image, fingerprint owner ID, fingerprint ID, and fingerprint representation as an entry in the database; Step 2: The basic service module accepts the fingerprint matching request from the client and sends the fingerprint image to be identified to the feature extraction module; Step 3: The feature extraction module extracts the minutiae of the fingerprint image to be identified and converts the minutiae into MCC codes; Step 4: The fingerprint retrieval module interacts with the feature value storage module to perform vector compression on the MCC code to make the code dense, and compares the feature value obtained in step 3 with the sample fingerprint feature value in the fingerprint library to obtain the ID corresponding to the fingerprint feature value with the highest similarity and its similarity score; The comparison is that the fingerprint retrieval module sets a threshold value for the similarity score. If the highest score exceeds the threshold value, it is considered that the retrieval fails and there is no matching fingerprint image in the fingerprint library. The vector compression uses matrix transformation to reduce the dimension of the original detail point code and binarize it at the same time, wherein the matrix used for transformation is obtained through learning, and a Laplacian matrix is used to maximize the correlation between the representation bits. Finally, each fingerprint detail point is represented as a 128-bit binary representation, and each fingerprint image corresponds to a binary vector set; Step 5: Return the fingerprint comparison result to the client that initiated the search request. If the search is unsuccessful, the result is returned as null, thus achieving fingerprint recognition. The fingerprint comparison result is the fingerprint image with the highest similarity in step 4 and its corresponding ID.
