| Face recognition,as a non-contact biometric recognition technology,has broad applications and has become a research hotspot in pattern recognition.Because the illumination has a great effect on face images in unconstraint environment,this dissertation mainly studies the face recognition technology under complex illumination conditions.To overcome the shortcomings of LBP and LTP in extracting feature face features,a distance-weighted Gaussian weighted local binary pattern(GWLBP)method is proposed to extract illumination-invariant features for face recognition.This method not only enlarges the scale of feature extraction by changing the weighting method,but also highlights the importance of the central pixel.The experimental results show that GWLBP outperforms LBP,LTP,AELTP and other feature extraction algorithms in terms of recognition rate performed on public face datasets.Besides working on feature extraction for face recognition,it is necessary to normalize the illumination of the original images and remove the illumination components.To overcome the shortcomings of Weber-Face and other algorithms on assumption that the local illumination is the same,two improved algorithms are proposed,namely,Gamma Transform-based Weber Face(?_WF)and Adaptive Weber Face(AWF).Gamma transform is used to improve the illumination condition on the whole image,which makes the assumption that local illumination is the same more rigorous,and then uses Weber-face to remove illumination.Moreover,the AWF method improves local illumination conditions by adaptive gamma transform at the pixel level,and hence remove illumination component effectively.The recognition rates of ?_WF+LBP on Extended Yale B and CMU-PIE databases are 99.32% and 99.75%,which outperforms the state-of-art methods.However,the recognition rates of AWF+LBP are 99.55% and 99.63% on two databases.The experimental results show that both ?_WF+LBP and AWF+LBP are top algorithms with high accuracy and strong robustness. |
PDF Full Text Request