Sparse Representation Image Classification Based On Multiple Residuals

Image classification is a technique for classifying an input image according to content,and is the core content of computer vision.In recent years,domestic and foreign scholars have proposed various classification methods and applied them to related fields.The image classification method based on sparse representation theory shows strong robustness.This method classifies the signal by selecting a suitable dictionary and classifying the residual by calculating the residual.In this paper,from the two aspects of the traditional sparse representation classifier and the sparse representation classifier with multiple residuals,the sparse representation classification algorithm is theoretically explored and experimentally verified,and some improvements are proposed for the shortcomings of the traditional sparse representation algorithm.The specific work is summarized as follows:1.The principle of sparse representation classifier is introduced in the system.Various methods for solving sparse representation coefficients are studied.Principal component analysis(PCA)and linear discriminant analysis(LDA)were used to reduce the data.HOG,Dense-SIFT and Gist were selected for feature extraction.Construct a sparse representation classifier for comparative experiments and compare and analyze the experimental results.2.Due to the limitation of the traditional sparse representation classifier,the text proposes a sparse representation face recognition algorithm with multiple residuals based on the traditional sparse representation classifier.The known residual weights are iterated by the control threshold.The multi-residence approximation is performed in the space of all possible results.The superiority of the algorithm is verified by comparison with the classical face classification algorithm.3.A multi-sparse classifier fusion image classification method with adaptive adjustment weights is proposed.By combining the information of multiple sparse representation classifiers,an adaptive classification model is generated to classify the images.Firstly,three sets of different features are extracted from the original image,and the respective sparse representation classifiers are trained.Then,according to theaccuracy of each sub-classifier,the final weight of each classifier is determined adaptively by iterative calculation.Finally,the output results of each sub-classifier are combined to judge the final category.The image classification and comparison experiments under single feature and the comparison experiments based on different classifiers show that the method has higher accuracy than the traditional method.