Research On Image Classification Algorithm Based On Dictionary Learning

With the development of science and technology and the progress of the Internet,various data have shown a sharp increase.As a form of intuitively expressing data,images have already penetrated into all walks of life in society.Facing the current massive image data,how to use effective classification technology to classify and manage it is of great significance.As an important technique in signal processing,sparse representation has aroused great interest of researchers in the field of image classification.Dictionary learning is an important step in sparse representation and a high-performance dictionary can obtain good image classification results.This thesis has carried out a series of research work on the related algorithms of the dictionary learning model.The main research contents are as follows:1.Aiming at the problem of how to improve the discrimination abilities of dictionary matrix and coding coefficient matrix,this thesis proposes a multi-sample dictionary learning algorithm combined with local constraints of Profiles.In order to improve the discrimination performance of the dictionary,the training samples are used to calculate the virtual training samples,and all the training samples are involved in the training of the dictionary.At the same time,the noise constraints are added to reduce the noise interference in the samples.In order to improve the discrimination performance of the coding coefficient matrix,the transpose matrix of the coding coefficient matrix(Profiles matrix)matrix is used to construct graph Laplacian matrix to describe the structural information of the coding coefficient matrix and dictionary atoms are used to measure the similarity between profiles and to construct Profiles local constraints.In addition,the proposed algorithm regularizes the coding coefficient matrix and noise matrix using 2l norm,which can effectively reduce the complexity of the algorithm.2.Aiming at the problem of how to use the trained dictionary to effectively coding the test samples to improve the classification accuracy,considering that the local constraint coding algorithm has good local information retention ability during coding step,and the histogram intersection method can effectively measure the similarity of data,so this thesis first uses the histogram intersection method to calculate the local information between dictionary atoms and test samples,and then uses the local constraint coding algorithm to code the test samples to achieve the improvement of some dictionary learning algorithms.In order to verify the effectiveness of the above algorithm,this thesis conducts simulation experiments on the public databases and compares the experimental data with related dictionary learning algorithms.The experimental results show that the algorithms in this thesis can obtain better classification results.