Research On Face Recognition Algorithm Based On Gabor Filter And Subspace Analysis

Automatic Face Recognition Technology(AFRT)have attracted comprehensive attention in the field of civil security,public safety,entertainment,customer service and the subjects field of image processing and pattern recognition.Facial expressions,age and other internal factors and the impact of external factors such as illumination,posture,occlusion,as well as twins,cosmetic and other factors,has brought great difficulties to face recognition.At present,the level of AFRT can't meet the requirements of many practical applications,so more innovative researchs need to be carried out continuously.A complete automatic face recognition system(AFRS)mainly includes the following process: Image Acquisition,Face Detection and Orientation,Imager Preprocessing,Feature Extraction and Classification Recognition.In this paper,the core steps of feature extraction of AFRS is taken as the starting point,carried out the thorough research of face recognition algorithm.Based on the research of the existing algorithms,the fast null space linear discriminant analysis(FNSLDA)algorithm and fast complete linear discriminant(FCLDA)algorithm have been proposed and merged with Gabor filter.The main work of this paper is as follows:(1)Discusses the research status and the existing problems of AFRT and its significance,analyzes the main research contents and the common methods of AFRT currently.At the same time,this paper also analyzes the evaluation criterion of AFRS,thus,the effectiveness of the algorithm has been analyzed comprehensively to make it closer to practical application.(2)Discusses the principle of Gabor filter and the process of extracting face features of Gabor,analyzes the meaning of the parameters and selection basis of 2D Gabor filter detailedly,and proposes the methord of using FFT to speed Gabor face feature extraction.Finally,the corresponding experimental analysis of the parameters selection of the Gabor filter is done.(3)The small sample size problem of the linear discriminant analysis(LDA)algorithm and two algorithms,linear discriminant analysis(NSLDA)and complete linear discriminant analysis(CLDA),for solving the "small sample" problem are analyzed in detail.And two algorithms,FNSLDA and FCLDA,are proposed to lift time performance in this paper.The involved theorems of these algorithms are proved compeletly.Finally,the time complexity of the algorithm is analyzed and verified by experiments,which verifies the effectiveness of the proposed algorithm in terms of time and recognition rate.(4)Based on the advantages of FNSLDA and FCLDA to reduce the feature dimension,the face recognition algorithm combining Gabor filter with FNSLDA and FCLDA is proposed.Then the fusion algorithm is used for the experimental comparison and analysis of pattern recognition and verification mode on the ORL face database.The results show that the subspace algorithm fused with Gabor filter has a good recognition effect.