Research On Palmprint Recognition Based On Gabor Features

Traditional identifications based on password have many drawbacks, such as the weaker security, being easy to be forgotten. For this reason, typical biometrics identifications mainly using fingerprint and face appear and have been widely used. Among them, palmprint recognition has been paid attention by many researchers as a novel technique because of higher recognition accuracy, higher recognition speed and better user-acceptance. But until now, we have not found which palmprint recognition method is the best one. So it is necessary to study the better algorithm of palmprint recognition. In this dissertation, a palmprint preprocessing method and two novel palmprint recognition algorithms are studied.The main work in this dissertation is summarized as follows:1. Based on an improved 8-neighbors search boundary algorithm, the author presents a novel palmprint ROI (Region of interest) extracting method.2. To extract useful features of palmprint ROIs, a set of Gabor filters on various orientations and scales are designed.3. Two palmprint recognition algorithms based on Gabor features are proposed:(1) Palmprint Recognition based on Gabor features'selection by using AdaBoost method. First, the binary templates of palmprint images'features (Gaborpalms) are selected by using AdaBoost algorithm. Then match them by using shift-matching method.(2) Palmprint recognition using Gabor feature-based bidirectional 2DLDA(Two dimensional linear discriminant analysis). The author applies two kinds of 2DLDA—horizontal 2DLDA (2DLDA) and vertical 2DLDA (V2DLDA). By using them, two kinds of features of Gabor-based images (Gaborpalms):horizontal discriminant features mainly extracted by 2DLDA and the vertical features extracted by V2DLDA can be obtained. To fuse these extracting features together, a distance-based adaptive strategy is designed.Experimental results carried out on PolyU_Palmprint_Database show that the proposed methods are effective. The author hopes the work can provide some references for the related research work.