Retrieving image via the integration of semantic annotations and visual features

Nowadays tag-based image search on image retrieval systems like Flickr, Google does not provide the option of relevance-based ranking. In this paper, a relevance-based ranking scheme which integrates both the visual features and the semantic annotations is proposed. The semantic correlation between database images and input tags is described using an improved form of Google distance, and the visual consistency between images is computed with Gaussian function. Experimental results demonstrate that the proposed approach taking into account the visual features and semantic annotations helps to improve the retrieval performance significantly.