Research On Face Recognition Algorithm Under Complex Illumination Conditions

Compared to other biometrics,face recognition is more convenient,friendly and direct,which is one of the research hotspots in pattern recognition field.Though face recognition has been widely put into practical application,several factors such as posture,expression,obstructions and illumination still have influences on recognition precision of face recognition system,among which illumination variation have more serious effects on recognition precision.This thesis researched methods of illumination invariant feature extraction for solving this problem,the specific works were as follows:(1)Weber-Face Algorithm based on partition was proposed.For Weber-Face Algorithm was to take feature extraction based on the whole gray image,which maybe could ignore local important facial feature information.To overcome this problem,an improved algorithm based on Weber-Face was proposed.Firstly,face image was partitioned and each sub-block was processed by Steerable Filters for reducing the effects of shadow edge in face image;secondly,each sub-block was processed by Weber-Face and a new face image was formed by connecting with all sub-blocks according to a certain order,finally,Non-negative Least-squares was applied as classification algorithm.This experimental results on Extended Yale B and PIE face database indicated that the proposed method still could obtain higher recognition rate under complicated illumination conditions,and possess better illumination robustness.(2)An improved Adaptive Smoothing Retinex Algorithm was proposed.Under complicated illumination conditions,adaptive smoothing Retinex filter can reduce the effects produced by side-illumination,and obtains good recognition rate,but the processed gray face images still troubled by shadow edge and overexposure.In order to reduce the influences of the above questions to face recognition rates under complicated illumination conditions,Adaptive Smoothing Retinex Algorithm was firstly used to deal with gray face image,then Weber-Face Algorithm was applied to solve overexposure and further reduce the effects of shadow edge.Finally,Linear Regression Classification Algorithm was applied to classification.The experimental results on Extended Yale B and PIE face database showed the proposed algorithm is of better recognition rates under complicated illumination conditions.