Content Based Image Retrieval via Sparse Representation and Feature Fusion

During the last two decades, content-based image retrieval (CBIR) has been widely studied. The limitations of low-level feature representation of images have been a thorny issue in image retrieval problems. In this paper, a CBIR algorithm based on sparse representation and feature fusion is proposed, in which global features and local features are combined to retrieve the images. Firstly, the GIST features are used to roughly retrieve the images with similar scene information by measuring Canberra distance. Then, a representation of the local features extracted from the rough retrieval results is obtained by sparse coding and feature pooling. Finally, Euclidean distance is used to measure the similarity of the sparse feature vectors to obtain the retrieval results. The proposed method is tested on the Coil20 and Caltech256 image datasets. The experimental results demonstrate that the proposed algorithm can achieve comparable superior retrieval performance to the existing single feature-based image retrieval algorithms.