Image Classification Based On Sparse Representation

With the development of multimedia technology and the needs of daily life, there are a large number of images are generated. At the same time, along with the popularization of the technology of image processing and computer network, the transmission speed of image is also improved greatly. On the other hand, images have been used in many fields. How to manage such huge amount of images is becoming more and more important. Image classification technology is an important way to solve this problem.This paper analyses the key technology of image classification based on sparse representation. The main work of this paper is as follows:1. Based on the analysis of image classification status and existing problems, the theory of sparse representation is introduced. Based on the analysis of the sparse representation models in the paper, the key of the models is discussed in detail which includes sparse decomposition and dictionary learning. Then, we give a brief summarization of the applications of sparse representation in the field of image processing.2. Based on the analysis of the traditional image classification framework, a new image recognition scheme is introduced based on ROI(region of interest) and sparse representation in the paper. Firstly, the ROI of the image is extracted and used as image feature. Then, the spg-lasso algorithm(a typical scheme of sparse representation) is adopted to realize the image classification. Experimental results tested on the Corel database prove the better performance of the new method.3. A novel image classification scheme is presented based on Gabor transform and sparse representation. The features of the training images extracted by Gabor transform are used as the dictionary of the sparse representation. And then, the test image is described as linear combination of the dictionary. Then the sparse coefficient of the test images and the reconstruction error are used for image classification. Experimental results tested on Corel database and two widely used texture databases show that the proposed method is better than the traditional methods.