Image Classification And Clustering Based On Sparse Representatinn And Block Diagonal Regularization

In the big data era,with the rapid development of artificial intelligence,sparse representation and dictionary learning have been successfully applied to encode dense data and facilitate image classification.Although the existing methods have been proved to be effective image classification methods,they are all looking for the most sparsest representation in the dictionary for each sample,ignoring the structural information between the samples.Although these shallow dictionary learning methods can learn strong discriminative dictionaries for specific categories of samples,the information contained in the shallow dictionary structure is far from enough for the classification task when the data dimension is too high or the number of samples is too large.It is not more accurate to make each type of sample reconstruct well from the same specific type of sample,and perform poor reconstruction from other types of samples.This paper completes three tasks to solve the above problems:1.In this paper,a new supervised dictionary learning method based on the priori of block diagonal phenomenon is proposed.A block diagonal regularizer is added to the affinity matrix to make the sparse representation matrix have an approximate block diagonal structure,which makes the learning dictionary more discriminant and more suitable for the classification task.2.As in work 1,the pixel information of the original image is directly used as the original dictionary for input,and the deep features of the image are not utilized.Therefore,deep learning method is used to extract deep features of images,and input them as the original dictionary to train the dictionary.Make the resulting dictionary more discriminant.3.In the previous clustering algorithm,the initial training samples are all disorganized,so it is not difficult to use the block diagonal regularizer to restrain.Therefore,in this work,we first sort the initial samples by using the perturbation matrix,and then use the block diagonal regularization,and get a good clustering effect.