A Unified System for Web Personal Image Retrieval

Although Web search engines have developed so fast, Web image retrieval is still a challenging task which needs more efforts to be made. In this paper, a system for personal image retrieval on the WWW is proposed. Particularly, a powerful text-based search engine (for instance, Google and Baidu, etc.) is first borrowed to start the initial retrieval with a given personal name. Then, the visual confidence with respect to somebody to be searched is explored (facial visual information in this case). By the analysis of the retrieved Web pages, the text-based semantic information is also extracted. With the fusion of visual confidence and textual information, the well known semantic gap can be narrowed down to some extent. The experiment results show that the proposed system achieved much better performance than the initial Web image retrieval via search engine used popularly.