Face Recognition Algorithm Based Onsingle Sample

Face recognition,as one of the most commonly used methods of identification,has a wide range of applications.In practical application,due to the limitation of human face image acquisition is affected by various factors,each and every one is often only a small number or a single image as the training sample,which can lead to the traditional variety I face recognition method to extract features of incomplete,and because a single training sample making one sample face recognition methods such as illumination changes,expression,block effect of robustness is low,low recognition rate.In view of the low face recognition rate of single sample,this paper studies the face recognition of single sample from two parts: sample expansion method and feature extraction method.Based on image subpattern,a feature extraction method combining LBP and HOG is proposed: S-LH method.Finally,in order to make up for the low robustness of S-LH for occlusion and further improve the face recognition rate of single sample,this paper combined this method with the sample expansion method and proved the effectiveness of the combined ES-LH method through experiments.The main work of this paper has the following points.(1)Study of existing sample expansion method,analyzes the characteristics of different samples of expansion method,try to different samples of expansion method,and to combine different methods to generate virtual samples as new training set respectively,and then respectively in ORL face database and FERET face database using the ESRC,FLDA,DMMA and depth of the SDA and traditional learning methods in combination with the new training set time experiment,the experimental results show that the paper proposed a method of the concentration of sample expansion method as a training set by the virtual samples samples can obviously increase the single face recognition rate.(2)To solve the problem of low recognition rate of single sample face recognition,this paper proposed lbp-hog fusion feature method based on image sub-mode: s-lh method.Traditional local binary pattern(LBP)to extract image texture feature for information cannot effectively describe the outline of the image,and because the single feature in the single sample face recognition and cannot accurately describe the original face feature information,so this paper tries to combine HOG algorithm to increase the description of the image contour information,at the same time using fused LBP-HOG feature and image sub model method to increase therobustness of interference information,such as illumination change,the facial expression change,etc.Finally,the effectiveness of the S-LH method is proved by comparing the ORL face database and AR face database.(3)Due to the S-LH method proposed in this paper to hide partially robustness is low,so in order to better improve the single sample face recognition rate,this article combines the best sample is obtained by chapter ii expansion method combined with S-LH method,using the generated virtual sample to form the new sample training set to increase the robustness of shade,finally through the four faces in public libraries and self-built carries on the experiment in a small library faces,and compared with different types of common method,the experimental results show that the proposed ES-LH method can improve the single sample face recognition rate.