Research On Face Recognition In Complex Conditions Based On Sparse Representation

Due to the rapid development of the face recognition,face recognition technology has made some progress even in complex conditions.The performance of face recognition algorithm is often seriously affected by environment interference,such as illumination,posture,occlusion,and expression.For solving the problem that the face recognition rate is declining in the complex environment,this paper proposed a novel descriptor named Enhance Center Symmetric Local Binary Pattern(ECS-LBP).By estimating on a series of extreme lighting conditions,the experiment results of the proposed descriptor confirm its robustness against illumination changing.Because the descriptor exist relatively poor performance in overexposure cases,this paper proposed a sparse representation with dual-features(SRDF)algorithm to supplement decision information.This way,the proposed SRDF algorithm overcome the shortcomings of decline of face recognition rate caused by occlusion and expression.In this paper,the LC-KSVD classification algorithm is improved to fuse the principle of feature classification at the decision-making layer to obtain the final result.By validating in Extended Yale B,CUM PIE and AR face database,the proposed algorithm has high recognition performance and good robustness in complex environment.In view of the fact that face recognition is easy to be affected by illumination,occlusion and other factors in complex environment,this paper proposes an effective and feasible idea.1)Proposed a ECS-LBP descriptor to encode image texture for reduing the serious effects of illumination.Through the fusing high frequency and low frequency details by wavelet transform,this algorithm obtain strong texture characteristics,which own stronger discriminative in serious illumination environment.Experimental results show that the descriptor has good robustness in serious illumination.2)With the deepening of the research,the improved feature descriptor in the case of excessive exposure exist relative performance degradation when compared with to the original algorithm in experiment.The research indicate that same kind of samples generate similar in sparse representation.To improve sparse representing method in recognizing part,this paper proposed an improved dual-features discrimination algorithm in the optimization process of the classification decision.Compared with the direct utilization of original feature descriptor,the proposed algorithm compensate the serious illumination,occlusion,and face expression,which always cause inadequate local feature and poor recognition performance.This way,the face recognition rate improve significantly rise.3)Based on a lot of comparison experiment in the Extended Yale B,AR,CMU PIE,controlled experiment is divided into two categories : illumination-expression and occlusion-illumination.Through the comprehensive comparison,the descriptors and dual feature discrimination algorithm proposed in this paper obtain effectiveness result under the complex environment,such as illumination and occlusion.