Research On Face Recognition Method Under Complex Illumination

With the development of the field of pattern recognition,the research on face recognition has made great progress.However,complex illumination conditions are still one of the key issues affecting the performance of face recognition system.Therefore,how to solve the influence of illumination variation on face recognition performance has become a challenging problem in practical face recognition technology.In this paper,the problem of face recognition under complex illumination conditions is the main research object,focusing on the in-depth study of normalization of illumination algorithm and illumination invariant feature extraction algorithm,and the corresponding solutions are proposed: self-quotient image face recognition algorithm based on improved Gabor and feature fusion face recognition algorithm based on illumination preprocessing chain.The specific research work in this paper is summarized as follows:(1)Aiming at the problem of illumination variation,this paper firstly studies the traditional image enhancement algorithm and the illumination invariant extraction algorithm based on the Lambert model.By analyzing the processing effect map and recognition accuracy,it is concluded that the traditional image enhancement algorithm can reduce the influence of illumination to a certain extent,but the recognition rate will decline sharply under the condition of complex illumination.On the contrary,the illumination invariant extraction algorithm based on the Lambert model can remove the illumination effect better,but there is still a large space for improvement.(2)By improving the Gabor self-quotient image model,a face recognition algorithm based on improved Gabor self-quotient image is proposed.Firstly,the improved weighted Gabor filter is used to extract smooth Gabor features from face images;Secondly,the illumination invariant features of images are obtained by using self-quotient image method;Thirdly,histogram truncation is used to normalize the selfquotient image;Finally,experiments are carried out on Extended Yale B and CMU PIE face databases by using the nearest neighbor method based on Pearson correlation coefficient.The experimental results show that this algorithm can greatly improve the face recognition rate compared with the traditional algorithm.(3)Aiming at the problem of low recognition rate of 2DLDA and LBP feature fusion face recognition algorithm under complex illumination conditions,a feature fusion face recognition algorithm based on illumination preprocessing chain is proposed.Firstly,the illumination pre-processing chain algorithm is introduced to perform illumination pre-processing,which solves the poor local robustness of 2DLDA caused by illumination variation;Secondly,the pre-processed images are separately subjected to block LBP feature extraction and 2DLDA feature extraction;Thirdly,the feature fusion of 2DLDA and LBP is performed by weighted fusion;Finally,the nearest neighbor classifier is used for classification and recognition on Extended Yale B and CMU PIE face database.Considering the problem of poor anti-noise ability of LBP features and sensitivity to smooth weak illumination gradient,a face recognition algorithm based on 2DLDA and LTP feature fusion of illumination preprocessing chain is proposed.The experimental results verify the complementarity of the two features and the effectiveness and robustness of the proposed feature fusion algorithm.