A Novel Method Of Image Sparse Decomposition And Image Retrieval

The image retrieval system based on image content completed image retrieval by extracting the visual features at the bottom of the image and matching these features. But it only extracted some of the features of an image, so that it lost a lot of information and the robustness of the image retrieval system became very low.The classification algorithm based on sparse representation was more and more popular with researchers and had high value in the field of machine learning. The greatest feature of this theory was not required to extract visual features of image. The complexity of classification algorithm based on sparse representation was very low and easy to implement.This paper proposed a staged image retrieval system based on image content and sparse representation. The main work and innovations of this paper was as follow:(1) This paper analyzed the image sparse representation theory and proposed adaptive step matching pursuit algorithm based on the traditional matching pursuit algorithm in order to complete image sparse representation. The accuracy of the new algorithm has improved and the complexity has reduced. So the new algorithm is a more effective method.(2) This paper analyzed the classification algorithm based on sparse representation and proposed the adaptive weight SRC under the image is missing or obscured. Then this paper completed the improvement of SRC algorithm through the step adaptive matching algorithm. The new algorithm improves the performance of image sparse representation.(3) This paper introduced the methods of image feature extraction and similarity computing methods. Then it represented the image by multi-feature based on the traditional image content. In order to strengthen the robustness of image retrieval system, considering the image content and SRC, we proposed a new image retrieval system and introduced the framework of the new algorithm. Next we realized the new algorithm on the basis of classification. At last, we did some experiments to verify the image system.