Exploiting Image Contents in Web Search

Web search is a challenging task. Previous research mainly exploits texts on the Web pages or link information between the pages, while multimedia information is largely ignored. This paper proposes a new framework for Web search, which exploits image contents to help improve the search performance. In this framework, candidate images are retrieved at first by considering their associated text information. Then, images related to the query are identified by analyzing the density of the visual feature space. After that, an image-based rank of the Web pages is generated, which is combined with the traditional keyword-based search result to produce the final search result. Experiments demonstrate the promise of the proposed framework.