Robust Face Recognition Algorithm Of Sparse Representation Based On Dictionary Extended

Due to society attaches great importance to the safety and convenience today,fingerprint,iris,and face recognition is gradually entering the application.The facial recognition is more friendly and convenient,so it becomes the focus on the biological recognition.Moreover,the facial recognition which is based on sparse representation is attracted much attention because of many advantages.On the relevant domestic and foreign scholars study basis,this paper is divided into three chapters to improve the face recognition algorithm which is based on sparse representation.Firstly,in view of the ordinary training sample dictionary learning only contains the global information in sparse representation-based classification(SRC),we introduce the atom dictionary which is related categories,and propose atomic and molecular dictionary joint extended weighted sparse representation of the human face recognition algorithm.Firstly the labeled atom dictionary is learned by each kind of training samples with PCA respectively,and at the same time the molecule dictionary is learned from training samples.So the extended dictionary is obtained by the combination of atom dictionary and molecule dictionary.When the input is tested,computing the weight according to the distance between the test sample and extended dictionary,so we can get the restructured dictionary associated with the current test sample.Finally,the test sample is classified by using of residual error SRC criterion.Secondly,the employed dictionary plays an important role in sparse representation classification,however how to build the relationship between dictionary atoms and class labels is still an important open question.Many existing sparse representation classification dictionary models exploit only the discriminative information either in the representation coefficients or in the representation residual,which limits their performance.To address this issue,we introduce a novel dictionary building method which is constructed by two parts: the common molecular dictionary and the discriminative atom dictionary.More specifically,the discriminative atom dictionary builds its relationship to class labels and the extended molecular dictionary can reduce the representation residual for all the classes.Since the new dictionary not only has correspondence to the class labels,but also has the perfect representation ability.Besides,the maximum probability representation is used for the final classification.In conclusion,the sparse coefficient of our method is sparser than the SRC,and our method can achieve the better performance.Finally,the image low-rank structure was destroyed due to both training samples and query sample always filled with lighting and uncorrupted,regularized robust sparse representation based on supervised low-rank subspace recovery for pattern classification is proposed.Firstly,using all training samples to form a data matrix,we can decompose the matrix as low-rank class-specific structure,low-rank non-class-specific structure and sparse error structure by the supervised low-rank decomposition.Then we apply PCA on the low-rank class-specific structure to obtain the transform matrix;and then using the transform matrix to project training samples and query samples onto low-rank subspace;Finally,we can apply weighted classification based on regularized robust sparse coding to classify the query image.