Image retrieval using semi-supervised learning

This paper proposes a novel scheme for the task of image retrieval based on one semi-supervised learning strategy. First, a pre-processing is utilized to tackle the problem of large computational cost involved in a large image database. Then, the similarity between the input query image and the remaining relevant images are measured to obtain initial relevance score. Finally, a semi-supervised learning algorithm, random walk and restart, is utilized to refine candidate ranking to improve the retrieval accuracy. Experiments conducted on a typical image database demonstrate the effective of the proposed scheme.