CN-115761817-B - Deep learning-based fingerprint image identification conversion method

CN115761817BCN 115761817 BCN115761817 BCN 115761817BCN-115761817-B

Abstract

The invention discloses a deep learning fingerprint image recognition conversion method, which relates to the technical field of fingerprint image recognition conversion and comprises the following steps of extracting skin fingerprint characteristics of a fingerprint image, recording fingerprint data of the fingerprint image, performing an image quality training set, performing characteristic information analysis on an fuzzy fingerprint image training set, performing a quality evaluation test after characteristic comparison, and analyzing and comparing characteristics of the fuzzy fingerprint image training set with characteristics of a clear fingerprint image training set. According to the deep learning fingerprint image recognition conversion method, characteristic information analysis is carried out on the fuzzy fingerprint image training set, characteristic information optimization of the fuzzy fingerprint image training set can be achieved through analysis and comparison of big data, and then the characteristic information in the image quality training set is combined, so that a fingerprint image database can be built quickly and effectively, and the fingerprint recognition conversion efficiency of equipment can be guaranteed.

Inventors

CAI FAN
ZHU TONGBO
LIN MUQUAN
LUO RUIYING
WU ZHENGYANG
XU ZHENMING

Assignees

闽南理工学院

Dates

Publication Date: 20260508
Application Date: 20221102

Claims (8)

1. A fingerprint image recognition conversion method based on deep learning is characterized by comprising the following steps: s1, extracting skin fingerprint characteristics of a fingerprint image; s2, recording fingerprint data of the fingerprint image, wherein the fingerprint data comprises position information of the fingerprint image and true and false information of the fingerprint image, when the collected fingerprint data does not accord with skin characteristics, the fingerprint image is a false fingerprint image, otherwise, the collected fingerprint data accords with the skin characteristics, and the fingerprint image is a true fingerprint image; S3, an image quality training set is used for screening and distinguishing the image quality of the collected real fingerprint images, the collected fingerprint images are divided into a fuzzy fingerprint image training set and a clear fingerprint image training set, the information of the fuzzy fingerprint image training set and the clear fingerprint image training set is marked through a marking module, and the information is stored through an information storage module; s4, carrying out characteristic information analysis on the fuzzy fingerprint image training set, marking and removing data which do not accord with characteristic information analysis data difference, carrying out training optimization on the fuzzy fingerprint image training set, and carrying out characteristic comparison on the trained fuzzy fingerprint image training set and clear fingerprint image training set information; s5, performing a quality evaluation test after feature comparison, performing a secondary fingerprint identification test on the data of the test passing through the fuzzy fingerprint image training set, and performing big data evaluation on the test result in a neural network analysis mode; S6, analyzing and comparing the features of the fuzzy fingerprint image training set and the features of the clear fingerprint image training set, extracting intersection data of the features of the fuzzy fingerprint image training set and the features of the clear fingerprint image training set, analyzing and storing the extracted intersection data, establishing an image training set feature library, repeating the steps, and perfecting the data of the image training set feature library; the characteristic information analysis comprises color space distribution values of all pixel points of the image, color space distribution values of all pixel points of the image and mean and variance of the color space distribution values of the pixel points of the image; the S6 further comprises the steps of analyzing and storing the extracted intersection data, screening and comparing the stored intersection data, and timely cleaning the compared redundant data.
2. The method for deep learning fingerprint image recognition conversion of claim 1, wherein the location information of the fingerprint image comprises: performing an image enhancement network on the live fingerprint image; the image-enhanced live fingerprint image is segmented into a plurality of live fingerprint enhancer sub-image blocks.
3. The deep learning fingerprint image recognition conversion method of claim 2, wherein the image enhancement network is realized by adopting a TP-GAN technology, and the low-quality fingerprint image marked by the image enhancement network is supplemented with the integrity and the definition according to the characteristics of the high-quality fingerprint image, so that the low-quality fingerprint image is converted into the high-quality fingerprint image, and the quality enhancement network is constructed.
4. The method for recognition and conversion of a deep learning fingerprint image according to claim 1, wherein the feature comparison values are color space distribution values of pixels of the image and mean and variance of the color space distribution values of the pixels of the image.
5. The deep learning fingerprint image recognition conversion method based on claim 1 is characterized by further comprising the step of utilizing Resnet as a base network to learn the characteristics of a clear fingerprint image training set, introducing big data information in the Internet and constructing a neural network.
6. The method for deep learning fingerprint image recognition conversion of claim 1, wherein the marking module is used for marking and classifying data of a fuzzy fingerprint image training set and a clear fingerprint image training set.
7. The method for recognition and conversion of a deep learning fingerprint image according to claim 1, wherein the intersection data is 70% -80% of the whole fuzzy fingerprint image training set data.
8. The method for recognition and conversion of a deep learning fingerprint image as claimed in claim 1, wherein the intersection data is 75% -85% of the whole clear fingerprint image training set data.

Description

Deep learning-based fingerprint image identification conversion method Technical Field The invention relates to the technical field of fingerprint image recognition and conversion, in particular to a deep learning-based fingerprint image recognition and conversion method. Background The types of biometric identification include facial, voice, iris, retina, vein, fingerprint identification, and the like. The fingerprint refers to the lines of the ridges and valleys on the skin of the front surface of the end of a human finger. Fingerprint recognition has become one of the most popular biometric methods at present, because fingerprint of everyone is unique and fingerprint is not easily changed with age or physical condition. With the continuous development of related technologies such as computer image processing and pattern recognition, the biological recognition technology is increasingly widely used. The fingerprint identification technology is one of a plurality of biological characteristic identification technologies, namely, the biological characteristic identification technology is to utilize the physiological characteristics or behavior characteristics inherent to a human body to carry out personal identification, fingerprint identification is to divide and compare fingerprints of identification objects to judge, the fingerprint identification technology is mature as one of biological characteristic identification technologies in the new century, the technology enters the production and living fields of human beings, the fingerprint identification technology is rapidly developed in recent years, the technology belongs to a relatively mature identification mode in a plurality of biological identification technologies, and along with the attack of hot tides of smart phones, the fingerprint identification technology is widely applied to the fields of smart phones, such as mobile phone unlocking, payment information, message confirmation and the like. In the prior art, in the actual fingerprint image recognition operation process, the image characteristic points are generally determined in a manual and mechanical combination mode, and the related image information cannot be determined in a device recognition mode under the condition that the image characteristic points are fuzzy, so that the phenomenon that the device cannot quickly and accurately recognize the fingerprint due to unclear fingerprints in the fingerprint recognition process of the existing device occurs, and the problem that the actual use effect of the device is poor occurs is caused. Disclosure of Invention (One) solving the technical problems Aiming at the defects of the prior art, the invention provides a deep learning fingerprint image recognition conversion method, which solves the problems mentioned in the background art. (II) technical scheme In order to achieve the purpose, the invention is realized through the following technical scheme that the fingerprint image recognition conversion method based on deep learning comprises the following steps: s1, extracting skin fingerprint characteristics of a fingerprint image; s2, recording fingerprint data of the fingerprint image, wherein the fingerprint data comprises position information of the fingerprint image and true and false information of the fingerprint image, when the collected fingerprint data does not accord with skin characteristics, the fingerprint image is a false fingerprint image, otherwise, the collected fingerprint data accords with the skin characteristics, and the fingerprint image is a true fingerprint image; S3, an image quality training set is used for screening and distinguishing the acquired real fingerprint images, the collected fingerprint images are divided into a fuzzy fingerprint image training set and a clear fingerprint image training set, the information of the fuzzy fingerprint image training set and the clear fingerprint image training set is marked through a marking module, the information is stored through an information storage module, the training efficiency of fingerprints can be improved through screening and distinguishing the false fingerprint images, and meanwhile, the images in the fuzzy fingerprint image training set can be trained through the arrangement of the image quality training set, so that the real training effect of the fingerprint images is guaranteed, the rapid training of the fingerprint images is realized, and the recognition conversion efficiency of the final fingerprint images is guaranteed; s4, carrying out characteristic information analysis on the fuzzy fingerprint image training set, marking and removing data which do not accord with characteristic information analysis data difference, carrying out training optimization on the fuzzy fingerprint image training set, and carrying out characteristic comparison on the trained fuzzy fingerprint image training set and clear fingerprint image training set information; S5, performin