Research On Face Recognition Under Different Illumination

As one of the most frequently used biological characteristics,face recognition has been widely applied in criminal investigation,public security,finance and other fields,was one of the most active research areas in computer vision and pattern recognition for years.Face recognition can obtain satisfied recognition results under controlled conditions.However,in real world,face recognition system have to deal with illumination change,which lead the recognition rate dropped significantly.Illumination is complex and hard to control,brings lots of adverse effects on face recognition system,such as image acquisition,feature extraction,classification and recognition.Theory and practice indicate that difference of the same person in different illumination can even bigger than the different person in same illumination;illumination variation still one of the biggest challenge in face recognition system.In order to solve the problem that the face recognition is greatly affected by illumination variation,this paper studied face recognition under different illumination in three aspects: illumination prepossessing,illumination invariant feature extraction and fusion rules.The main work is as follows:1.Study and improved the self quotient image method.In order to solve the problem that the self quotient image based methods ignores the selection of features,a novel face recognition approach based on self quotient image and random projection was proposed.Firstly,the self quotient image is used for illumination normalize in order to remove the effect of illumination.Secondly,constructs an initial sample space using linear discriminate analysis.Furthermore,by projecting the sample to many different subspace by random projection,the diversity and completeness of the sample feature are enriched,so as to increase the discrimination between different samples and improve separability.Finally,the nearest neighbor classifier is used to select the training sample with the minimum distance from the test sample to determine the category of the test sample.2.Study and improved a novel illumination invariant feature extraction method.Aiming at the problem that facial image can hardly describe accurately by single feature,proposed a face recognition method which integrates local sensitive histograms and Gauss of Laplace feature.This paper makes full use of the theory of locality sensitive histograms,according to the distribution characteristics of facial images,separate the gray value of the pixel in the face image to float point weight according to the distance between the pixel of the distance center,to reduce the background interference,in order to extracts illumination invariant features,meanwhile,Gauss of Laplace is used compensate for the loss of details,so as to realize more accurate describe of facial image.3.Study the method of feature fusion.In order to solve the problem of poor applicability of fixed weight,a method of using standard deviation in the feature fusion is proposed.In face recognition,the extracted features may contained noises,which will affect the correct classification in face recognition.In this paper,according to the standard deviation of the feature contained different useful information,by calculating the standard deviation of the features,the weight values in feature fusion are determined and the features are adaptive fusion.