Research On The Key Technology Of Face Recognition Under Unconstrained Environment

Face detection and recognition is an important research topic in various area.After several decades of research,the performance of face recognition has received more progress under contained environment.However,Face recognition performance usually drops heavily under unconstrained environment due to illumination variations,complex background and other factors.Hence,a novel face recognition algorithm under unconstrained environment is presented in this paper.Firstly,a illumination pre-treatment method based on image fusion using wavelet decomposition is explored,based on which the face detection is conducted.Experimental results have shown that it cannot only reduce the influence of lighting variance for feature extraction but also can increase the accuracy of face detection.Secondly,a novel feature extraction method based on CS-LBP combined with DBN is proposed.By fully using the extracted local textural features which is not sensitive to illumination variation,the proposed method can compensate the shortage that DBN usually ignore the more distinctive local features when using original face image,achieve more robust deep features.Thirdly,sufficient experimental results on FERET,YALE-B and CMU-PIE database have illustrated that the proposed method has a significant improvement on recognition rates compared with LBP,DBN and other local feature extraction methods combined with DBN when combating the influence of illumination variation and complex background.Finally,based on the above mentioned illumination pretreatment algorithm and face recognition algorithm,this paper proposes a real-time face recognition system which can complete the whole face recognition process.