Image Classification Algorithm Based on Sparse Coding

Abstract

In this paper, the sparse coding and local features of images are combined to propose a new image classification algorithm. Firstly, online dictionary learning algorithm is employed to train the visual vocabulary based on SIFT features. Secondly, SIFT features are extracted from images and these features are encoded into sparse vector through visual vocabulary. Thirdly, the images are evenly divided into I*I areas and the sparse vectors in each area are pooled, getting a fixed dimension feature vector which represents the whole image. Lastly, to achieve the purpose of image classification, we use support vector machine classifier for learning and recognition. Results from the Caltech-101 and Scene-15 data sets show that, compared with the existing algorithm, the proposed algorithm has a better performance, which can effectively represent the feature of images and improve the accuracy of image classification greatly.