Extraction Algorithm Of Face Invariant Feature Under Complex Illumination

The technology of face recognition has been used in many important occasions and is also a pivotal component of Machine Vision.However,the accuracy of the existing face recognition system highly depends on image quality.The performance of face recognition system will be degraded seriously when processing unconstrained face image with low-quality.The results of face recognition can be affected by many factors,and illumination is one of the important factors.In this paper,we will focus on the face recognition algorithm for preprocessing face under complex illumination in the face database via the following aspects:1)Investigating the image enhancement methods such as Gamma correction,histogram equalization and logarithmic transformation to normalize the image illumination based on EYaleB database.Theoretical analysis and experimental results show that,to some extent,the above methods can weaken the light,however,under the severe light interference,the improvement of visual effect by light normalization is limited,and the excessive smoothness or enhancement may occur.2)Based on the lambert reflection model,a correlation algorithm for image enhancement is proposed in this paper.The validity of the algorithm is verified by experiments,and the analysis shows that the image will produce "halo" phenomenon when using MSR algorithm for processing.In the SQI method,the size of the window and the scale of the filter core have impacts on the image processing,while the Multi-Scale SQI method can achieve better processing effect.3)This paper proposes a method of adaptive fusion of histogram equalization and logarithmic transformation based on standard deviation and mean.In order to solve the shortcomings of the mentioned traditional algorithms,this paper deals with gray-scale images based on image standard deviation and mean value,uses the standard deviation as a parameter to measure the threshold of image logarithmic transformation,and enhances the gray-scale image by limiting contrast adaptive histogram equalization,then obtainsthe mid-point of the data set by mean calculation as the weight of the combination of log transformation and AHE algorithm.Finally,the image can be enhanced while weakening the influence of illumination.4)A method based on wavelet transform and SQI theory is proposed to extract the illumination invariant features of image.Decomposing the image into low frequency and high frequency by wavelet transform.And two-sided filter is used to filter the low frequency coefficient to estimate the illumination information contained in the low frequency component,obtaining the Invariant Illumination features through Multi-Scale SQI.In order to estimate the illumination information contained in the high frequency coefficient of the image.Finally,the illumination information of the original image is estimated by the reconstruction of wavelet coefficients,and finally the image is enhanced by linear stretching.Compared with the other algorithms,the algorithm proposed by this paper can make full use of the features which can be used to identify from the high frequency and low frequency components of the image,and can achieve better results under the changing illumination conditions.