Research On Face Recognition Algorithm Based On Linear Representation

Recently,the face recognition technology as one of the most attractive research fields in computer vision and pattern recognition has been used in many real-world applications such as Access Control System and Public Safety-Visualization System.Various algorithms for face recognition have been developed from past to now.Representation based classification methods(RBC)are significant methods in the field of face recognition,because they are easy to operate and can bring about high efficiency.This dissertation proposed three modified linear representation based classification methods for face recognition.In this paper,we proposed a supervised dimensionality reduction(DR)algorithm which suits sparse representation based classification method(SRC)well and improves the discriminating ability in the low-dimensionality space.The proposed method utilizes the fisher discriminant criterion and low-dimensionality reconstructive restriction to extract the discriminating structure of data.Then,an improved collaborative representation classification algorithm is proposed by using the gradient image of the face.The improved algorithm can achieve better recognition accuracy by merging the original image and the gradient image.Finally,we propose a relaxed minimum squared error(RMSE)algorithm.This RMSE algorithm require the mapping transform all training samples into an approximate of the predefined class labels rather than the class label itself.The mapping successfully reduces the predicted errors of test samples.Experimental results show that the RMSE classification algorithm which transforms training samples into 'relaxed' class labels rather than predefined class labels can perform better classification.Extensive experiments about the three improved algorithms proposed in this paper in several face databases,such as ORL database,FERET database,GT database and so on can demonstrate their classification accuracy.Moreover,in each of the verification experiments,we compared the latest linear representation based on the classification algorithm with the proposed method in this paper.Extensive experimental results doubtlessly demonstrate that the proposed methods can effectively improve the classification accuracy.