Research On The Frequency Analysis And The Uncorrelated Fisherface Apporach

Biometrics is a kind of science and technology using individual physiological or behavioral characteristics to verify identity. It provides a highly reliable and robust approach to the identity recognition. Time-frequency analysis tools and Linear Discrimination analysis each is one of the important beaches of Biometrics.This dissertation mainly studies two kinds of time-frequency analysis tools: the Wavelet Transforms and the Fractional Fourier Transforms and the Linear Discrimination analysis in biometrics. The mainly contributions are as follows:(1) The theories and applications in Biometrics of the Wavelet Transform and the Fractional Fourier Transforms are introduced. Faced the problem of low efficiency of the Fractional Fourier Transform, A new approximate fast fractional Fourier transforms is proposed and testified that it is more quickly than the fractional Fourier transform which is computed by definition by theoretical and experimental provements.(2) Small size sample problem is the one of the most difficult problem in Linear Discrimination analysis. An Uncorrelated Fisherface method (UFA) is put forward. The method reveals the relationship between the Fisherface method and the Uncorrelated Optimal Discrimination Vectors method and also proves that the UODV method is theoretically superior to the Fisherface method. The experiments using the biometrics datasets proves that the new method has better performance than the Fisherface and UODV methods.(3) In order to select the"good"discrimination vectors for recognition, a new criterion of choosing the discrimination vectors is proposed. The less discrimination vectors are needed, and the recognition rate remains at the same level. Hence,it reduces the burden of computing.