Research On Locality-Sensitive And Combined Sparse Representation Based Face Recognition

As one of the most challenging research topics in the field of pattern recognition and computer vision, face recognition has attracted much attention with no need of object cooperation, remote covert operation and friendly identification process. And It has many applications in business and law enforcement besides its scientific significance, for instance, supervision, security, communication, human-computer interaction, etc. After 30 years of research, Numerous face recognition methods have been proposed by researchers and scientists.With the rising of Compressive Sensing, especially the core technology of sparse representation which can not only reduce the cost of data analysis and processing, but also improve the efficiency of data compression. The methods which based on sparse representation received extensive attention because of its excellent performance and robustness to noise and occlusion. So that researchers focus on the research of face recognition based on sparse representation to realize the high accuracy and improve the technology.This paper has carried on deep research on the face recognition method based on sparse representation. On the basis of this, the constraints of locality-sensitive, group sparsity and combined sparsity are added to the sparse coding, which can improve the classification performance of the face image. The main contents and innovations of this paper are as follows:1) This paper first introduced the current status of the field and its related technologies, then outlined the framework of face recognition based on the research methods of this paper, and summarized the main focus of this paper, that is the relevant basic knowledge of sparse representation.2) This paper presents a novel classification method with locality-sensitive sparsity and group sparsity constraints for robust face recognition. This method learns group sparsity and data locality at the same time. It not only takes into account the grouped structure information of the training data dictionary, but also integrates the data locality, thus can learn more discriminating sparse representation coefficients for face recognition. To testify the effectiveness of the method that we proposed, we perform experiments for face recognition on the ORL, AR, and Extended Yale B databases.3) This paper presents a novel classification method for face recognition which based on kernelized locality-sensitive group sparsity representation and extreme learning machine. First we implement both group sparsity and kernelized locality-sensitive constraints under the framework of sparse representation. This framework not only utilizing the grouped structure information embedded in the training data, but also retaining the locality constraints on sparse coding, and joining the Gauss kernel to improve the performance of the constraints. Then we combine this framework with the extreme learning machine, make up the accuracy limit of extreme learning machine which has fast learning speed in classification. Sparse representation which has better classification performance, however, is more time consuming. Through our method we can achieve a faster speed of face recognition, while maintaining a better classification performance. We have experiments on ORL, AR and Extended Yale B database respectively, to test the recognition rate of our method and the time of classification.4) This paper presents a novel classification method with non-cascade Gabor dictionary and combined sparsity constraint for robust face recognition. This method uses the non-cascade Gabor features with eight-direction and five-scale information to instead the original image features for sparse representation. It not only can overcome the interference of expressions, illumination, and pose etc. variations, but also could improve the classification performance via sparsity constraints. On the AR and Extended Yale B databases, we test the effectiveness of our method.