Research On Face Detection And Recognition Under Uncontrolled Environment

With the coming of the information society,online shopping,access control systems,electronic payments and other applications are more popular.At the same time,the requirement of safety is also more important.Face recognition has attracted more and more attentions because of its non-controllability,friendliness and low cost.At present,there is few face recognition algorithms for non-control environment.Because the face recognition under the non-controlled environment is disturbed by light,gesture,expression,age and ethnicity,the classical face recognition algorithm can not get the ideal recognition rate.To solve this problem,we carry out research on face detection and recognition.The content includes the following several aspects:(1)Based on the biological visual attention mechanism,we study face detection algorithm based on visual saliency under complex environment.Based on the GBVS algorithm,we detect the target face.Experiments show that the algorithm can not only get rid of the interference factors such as illumination,gesture and facial expression,but also can detect face accurately and without human intervention.On the one hand,it reduces the computational complexity,shortens the feature extraction time.On the other hand,the interference factor is removed,the feature extraction is more accurate and the recognition rate is greatly improved.(2)In order to realize robust face recognition under uncontrolled environment,the SRC algorithm is studied.However,the dictionary of SRC is constructed by pixel values,not only existing noisy interference and redundant information,but also can't represent the nature of picture.This paper proposes HOG_SRC algorithm.Experiments show that new dictionary contains more local gradient features,which can encode the test sample more accurately and improve the recognition rate.However,the running time is long.To reduce the running time,this paper introduces dimensionality reduction,which not only greatly reduces the running time,but also significantly improves the recognition rate.(3)Under ideal circumstances,the coefficients obtained by coding the test samples should have good sparsity,but the sparsity of the coefficients is very weak in practical applications.The reason is that dictionary atoms isn't highly independent,to solve this problem,this paper proposes PCA_SHIFT_SRC algorithm.Experiments on the LFW and PubFig databases show that the PCA_SHIFT_SRC algorithm does improve the face recognition rate.