Face Recognition Based On Sparse Representation

The application of face recognition technology is increasing in the society,such as face swipe card,face payment and even the cross-age recognition technology like searching for children missing for years have been applied.At present,face recognition is a hotspot in the field of pattern recognition and computer vision.However,there are many problems in face recognition algorithm,especially in the face feature extraction and expression,and it is very easy to be influenced by illumination,expression and posture.In recent years,a new research direction has been developed by the sparse rep-resentation classification for the classification algorithm whose excellent recognition performance has been paid attention by many researchers.Traditional face recognition based on space representation,the use of training datasets are often not extracted fea-tures,but directly translated from the original image data to a one-dimensional vector and inputted into the classifier,which can greatly affect performance of the sparse rep-resentation.The hidden information in the human face image is almost non-linear,and the conversion of raw data to one-dimensional vector may lose a lot of useful infor-mation.In this case,if sparse representation classifier is used directly,bad results will be get.In order to improve the performance of the sparse representation classifier,this paper studies the above problems.Firstly,the weighted sparse representation face recognition algorithm was re-searched and a new weighting matrix was designed to improve the weight relation-ship between the data.On this basis,the histogram of oriented gradient feature ex-traction was researched,then face recognition based on multi-scale hist of oriented gradients with weighted sparse representation(WHSRC)is proposed.In this algorith-m,a multi-scale gradient directional feature extraction algorithm is designed and used by the weighted sparse representation face recognition algorithm.The advantage of multi-scale gradient orientation is that the local texture information and global texture information can be very well characterized by the facial features,so it takes into account the global and local information of image data.Secondly,the research of self-encoding neural network is introduced to the sparse representation face recognition algorithm,and face recognition based on self-taught features with sparse representation(STSRC)is proposed.Compared with feature ex-traction based on image processing technology,data feature extracted by neural net-work is more effective.Using neural network,let the training data taught by itself,and finally the more abstract features are extracted.then I convert face data into the learning feature space,and use the sparse representation classifier to decide its class.Finally,we tested the two algorithms proposed.Experimental results shown that the proposed two algorithms can improve the recognition rate effectively.In addi-tion,the two algorithms proposed was compared separately,and the results shown that the WHSRC algorithm is suitable for high dimensional face recognition under simple scenes,and the STSRC algorithm is suitable for face recognition in complex environ-ments.