A Multi-Feature Fusion Face Recognition Method Using Fuzzy Clusteirng Analysis

This paper presented the development of face recognition, meanwhile introduced severalrelated algorithms. Then it demonstrated how to extract features from human faces torecognize the accurate identification. A method of face recognition using fuzzy cluster tocombine subspace and Gabor wavelet was presented.Generally speaking, there are several components of face recognition system, includingface detection, tracking, verification and recognition. The implementation of face recognitioninvolves the following phases: firstly extract features from the face region, then match themwith the sample database, figure out the exact class which the test image belongs to. Thecritical factor that affects the final result is whether we can get sufficient information topresent the human face. Subspace and Gabor wavelet are widely used in extracting andexpressing faces information. Subspace is used to transform the human face from a highdimension to a descriptive space, meanwhile compress the original data. While Gabor waveletexhibits spatial and orienting features in frequency domains. Moreover Gabor wavelet isrobust to illumination and facial expression. However subspace method has weakness inextracting detail information and adapting illumination variety. In the other hand, Gaborwavelet suffers from redundancy and complexity of computing and storing.In order to solve this kind of problem, a novel method was presented. Firstly it uses fuzzycluster to analyze the distribution of sample database in subspace. Then it classifies thesample database into two categories. One of them contains samples which separate fromothers. The other is constituted by samples that mix up with other samples. The presentedmethod deals with the first category by subspace method, while the second one by Gaborwavelet. The other innovation in this paper is that a multi-Gabor fuzzy weighted recognitionmethod is proposed. Instead of the traditional ensemble Gabor presentation, it uses features indifferent Gabor channel separately, in order to make use of the difference between channels.This paper compared recognition rates using subspace, Gabor wavelet and the presentmethod. The experiment shows the last one combines the advantage of subspace and Gaborwavelet, and it has benefit in classifying and is robust in complex environment. Meanwhileanother experiment on the efficiency of fuzzy weighted rules in multi-channel Gabor wavelet shows the present method has higher recognition rate. At last we compared the time and spacecomplexity of the tradition methods and the proposed one. Statistics data shows theinnovative method is more suitable for using.The proposed method has some contribution in feature extraction and time complexity inface recognition. Currently researches and applications about face recognition are mostlybased on a limited database. And the performance is affected by illumination, background,and gesture of the human face. Nowadays the amount of information is rapidly explodingthrough the internet. We are still facing a lot of challenges in the face recognition domain.