K-SVD Algorithm For Image Sparse Representation And Its Application In Face Recognition

In the fields of human-computer interaction and intelligent monitoring,face recognition has been widely applied as a key technology for identity verification.However,the accuracy of face recognition is reduced to a certain extent due to the influence of factors such as illumination,occlusion,expression and posture in the real environment,which makes the study of face recognition technology challenging.In recent years,the idea of sparse representation has been introduced into the fields of machine learning and pattern recognition due to the reliable statistical theory,and good results have been achieved in face recognition.The core idea of face recognition method based on sparse representation involves such influencing factors as optimal construction of over-complete dictionary,correct solution of sparse coding and appropriate selection of classification algorithm.This article introduces the face recognition method based on sparse representation.The specific work is as follows:The theoretical basis of the sparse representation algorithm and its relevant theoretical support in face recognition are emphasized.The general architecture of the image sparse representation algorithm is introduced in detail,including the sparse representation of the signal,the sparse representation model,the solution of the sparse representation problem,and the dictionary learning algorithm.In order to improve the accuracy of the face recognition method based on sparse representation,the K-SVD algorithm for sparse representation of images is mainly studied.Therefore,a face recognition method based on improved LC-KSVD dictionary learning is proposed.Firstly,PCA is selected to decompose the error items in K-SVD dictionary updating phase instead of directly performing SVD decomposition on the error items to update the dictionary atoms.Secondly,the improved process was applied to the dictionary update stage of the LC-KSVD algorithm,which improved the dictionary learning ability.Experimental results show that the performance of the proposed algorithm is relatively stable and the accuracy of face recognition is high.To deal with the influence of illumination change on face recognition accuracy,a face recognition method under illumination changes based on improved LC-KSVD dictionary learning is proposed.Firstly,the combination of histogram equalization andwavelet denoising is used to preprocess the training sample image,so that the illumination invariance description of the face image is obtained.Then the initial dictionary is constructed by using the dimension reduction performance of PCA method.Next,the initial dictionary is updated,and the improved LC-KSVD algorithm in the dictionary update stage is used to obtain the new dictionary with the ability of presentation and differentiation.Furthermore,the proposed method calculates the sparse coefficients corresponding to the feature matrix of the test sample image under the new dictionary,the characteristic matrix of the test sample image is reconstructed by quasi-association,while solving the corresponding reconstruction error at the same time.Finally,the classification of test sample images is realized according to the reconstruction error.The relevant experiments on the face database prove that the algorithm can improve the recognition accuracy to a certain extent,and better solve the impact of lighting on the accuracy of face recognition.