Palmprint Recognition Based On Feature Extraction

With the rapid development of biometric identification technology, people pay more and more attention to their self-identification. As an emerging biometric identification technology with the advantages of high recognition accuracy, short recognition time, high user acceptance and low-cost hardware, palmprint recognition technology has not been widely used. Therefore, studies on palmprint recognition technology will be a very promising subject.In this paper, studies based on the method of palmprint feature extraction are carried out, from four parts of pre-processing、regional positioning and segmentation、feature extraction and classification.In the pre-processing stage, palmprint images are denoised by using median filtering, and the contrast of palmprint images is enhanced and stretched by using Butterworth highpass filter.Firstly, palmprint images are done binary processing in the regional location and segmentation, the edge of palmprint is extracted based on Sobel edge detection operator, then the angular points of palmprint image are determined by using the method based on mathematical morphology, and the palm area of interest is segmented based on the angular points.In the feature extraction stage, two methods of palmprint feature extraction based on Gabor transform are analyzed and discussed. Based on Gabor transform, Gabor texture features of palmprint image(dimension of 128×128) are extracted with five scales and four directions, after dimension reduction by using Mean and PCA, a feature matrix of dimension of 640×8 is obtained. Based on improved Gabor transform, the dimensions of palmprint image are reduced by using Mean, and a palmprint image of dimension of 64×64 is obtained. Then palmprint image is divided into 64 sub-images after dimension reduction, Gabor texture features of palmprint image are extracted with two scales and four directions. Then feature weighting groups are divided according to the certain rules, weight-value of sub-images of Gabor feature within the feature weighting group is calculated. Finally, eight weighted feature sub-images of Gabor filter of each sub-image are equalized, and a characteristics matrix of dimension of 8×8 is obtained. Finally, a palmprint feature image of dimension of 64 × 64 is obtained from a palmprint image of dimension of 128×128. Feature matrices obtained by two methods will be used as sample characteristics of training set for subsequent classification.K-nearest neighbor is used to do classification and recognition in the classification stage. The Euclidean distance is calculated between the feature matrix of the sample of training set and the feature matrix of the test sample to find out its K nearest training samples. Based on statistical theory, the type of the majority of the training samples is the type of test sample.Palmprint images from palmprint database of biometrics research center of Hong Kong Polytechnic University are used to do test experiments in the system. Experimental results show that the palmprint recognition accuracy rate of extraction method based on Gabor transform is 96.67%, and the palmprint recognition accuracy rate of extraction method based on improved Gabor transform can reach 97.33%.