Handprint Recognition Based On Cpd And Feature-level Fusion

Biometric technology has been extensively studied internationaly, feature extraction is one of the most basic issue in biometric technology research. So far, feature extraction methods can be divided into linear and nonlinear feature extraction method. Although the existing linear feature extraction method have been successful used in the identification and image processing area, but they are essentially based on linear transformation,can not extract the image model of high-level information, therefore,can not express such as palm-prints, fingerprints and other images of the distribution structure of complex non-linear. Kernel-based feature extraction methods gradually became the focus of everyone's attention. In this paper, in-depth study of nuclear function has been made. This paper analyses the kernel function, and puts forward conditional positive definite function according to its own characteristics of hand image(palm-prints, fingerprint). Finally, this paper presents the kernel matrix fusion methods under relational measure framework by using multiple modality algorithm and feature-level fusion.This paper studies the following two elements:1. Palmprint and fingerprint analysis of the characteristics of their own to study the various existing problems in the nuclear function of the proposed terms of positive definite kernel (CPD). And the experimental conditions, based on positive definite proof of the hands of the nuclear pattern recognition algorithm with anti-pan and highlight the characteristics of local information, to a certain extent, improved the recognition rate.2. Feature-level fusion, this paper gives the KPCA-based multi-biometric recognition algorithms, including the right to conduct fingerprint and palmprint feature level fusion algorithm of nuclear matrix, and a variety of feature-level integration of nuclear function fusion algorithm. Experiments proved that integration was indeed able to achieve a more high identification rate and the authentication error rate even lower.