Research On Dictionary Learning And Its Related Algorithms In Image Classification

With the rapid development of network technology,humans entered the era of information and intelligent.The spread of network technology makes a large number of digital images through various channels emerge every day.The digital image is an important carrier of visual information,because it's more accurate and intuitive.However,in face of such massive image data,it's getting harder to find what we are interested in.Therefore,the image classification technology has attracted widely attention of researchers,and become one of the most important research contents in the field of pattern recognition and computer vision.In many image classification models,the spatial pyramid model is widely applied,it divides the image into a more sophisticated space area layer by layer,and then calculates its local feature histogram in the sub-region,showed a stable performance in the image classification.In view of the advantages of spatial pyramid model,this paper makes an in-depth study on the dictionary learning,feature coding and related algorithms of the image classification stage based on the basic framework of this model.The main contents are as follows:(1)In this paper,based on the intensive study of sparse coding algorithm,compared with traditational sparse coding,considering the advantages of elastic net,this paper combine it with Spatial Pyramid Matching model,and applied it to image classification,Experimental results vaidated achieved high classification accuracy.(2)Propose a new algorithm— Non-Negative Elastic Net Sparse Coding Algorithm.Analyzing and comparing the characteristics of non negative sparse coding and elastic net sparse coing algorithm,this paper introduce non-negative constraint to the objective function of elastic net optimization model.Compared with the traditional elastic net sparse coding algorithm,the proposed algorithm not only can introduce the prediction capability and effectiveness of the coding,but also make the coding of similar feature descriptor similar,increasing the stability of the model.(3)The proposed algorithm combine with Spatial Pyramid Matching model is applied to image classification.Experimental results vaidated that the proposed method achieve better classification accuracy than existing methods.The experiment of this paper is based on the basic framework of the SPM model,and the LibSVM Package is used in the feature extraction phase.In order to verify the validity of the proposed algorithm,we have done a lot of experiments on the dataset of Caltech-101,Caltech-256,Scene-15 and UIUC-Sports.Compared with the classification method based on sparse coding,the experimental results show that the proposed model based on non-negative elastic net sparse coding has higher classification accuracy.