| Near-infrared palm vein recognition technology is a new technology.In recent years,with its unique advantages,it has risen rapidly in market,which attracted widespread attention of domestic and international researchers.In this thesis,through the system design,algorithm design and hardware implementation,a complete near-infrared palm vein recognition system is developed.There are two valuable works in the system algorithm as follows.At first,in the image pretreatment,taking the median point as stable point in linear stretching method is used in the contrast enhancement of near-infrared palm vein images.A good contrast enhancement algorithm has an important affect on the performance of image feature extraction.After consulting the data,it is found that most of the existing techniques use some conventional algorithms such as adaptive local contrast enhancement algorithm,histogram equalization algorithm and platform histogram equalization algorithm.Due to the fact that the palm vein image has a very concentrated histogram distribution and the contrast is very small,it is easy to lose some image information by these three methods.And the histogram distribution of the whole image can not be stretched to the whole gray level evenly.By comparing the histogram,we can find that this method is very uniform in nearly 85% of the entire gray level,and it hold the information of image very well.Secondly,the RANSAC algorithm is uesd in the image feature extraction and image feature matching.This could improve the accuracy of system matching very well.In the investigation of dozens of feature extraction algorithm,taking the factors of real-time,robustness and recognition performance into consider,the SIFT algorithm has been choosen to extract image features.This algorithm is very good in the different environment.And it has remarkable effect on the near-infrared image feature extraction.Since the palm vein image's details are not sufficient,and the extracted features is not so independent.It must result in a bad effect in image feature matching with traditional SIFT matching method.In the legal matching,the false matches is close to 30%,while in the illegal matching,a lot of false matches are judged to be correct match.To solve this problem,RANSAC algorithm is used in near-frared image feature matching.I use this algorithm to find the correct mathematical model of all matching pairs.Finally,more than 70% correct matching pairs are selected in the legal matching,and nearly 90% of the false matching could be eliminated in the illegal matching.This algorithm improves the accuracy of system matching.In the hardware implementation,this thesis gives the parallel structure design of the SIFT and image feature matching algorithm on FPGA,and also gives a detailed description of the specific design process.The results of simulation in the design process has been compared with the results of MATLAB fixed-point test.Finally,the corresponding hardware test based on the Z-turn Board TOP development board has been made.In a word,I use the algorithm of taking the median point as stable point in linear stretching,the SIFT algorithm and the RANSAC algorithm to solve the problems in image pretreatment,image feature extraction and image feature matching.In this thesis,through the works of the system architecture design,algorithm design and the final implementation of FPGA,a system with strong robustness and 90% correct recognition rate was designed.It overcomes the disadvantage of the insufficient information in the near-infrared palm vein image,and it also has high practical value.The whole system algorithm not only has the practical value in the near-infrared image processing,but also has certain reference value for many field,such like the pattern recognition and the Internet of things.Through the whole works of the system architecture design,algorithm design and the final implementation of FPGA,I improve the ability to think systematically,analyze and solve problems independently. |
PDF Full Text Request