LiveImage: Finding Web Images by Relevant Concepts
Search Results for the query 'Shanghai'
The increasing demand for high-performance Web image search engines is being driven by the rapid growth in the number of Web users and in the availability of image collections on the Web. It's very easy to retrieve a very large number of images by submitting a short query to a Web search engine. Unfortunately, the returned images are not well organized in a conceptually meaningful way. To help users find images in a conceptually simple way, we propose an approach for organizing retrieved images based on auto-generated relevant concepts. The images listed in the retrieved results are organized according to their similarity to the relevant concepts, which makes it easier for users to locate images of interest with the corresponding concept names and relevant images.
Our goal is to discover meaningful concepts relevant to Web image queries. Collecting and organizing such relevant concepts manually is infeasible due to the dynamic nature of the Web environment. We are interested in developing an automatic approach that organizes users' query terms into auto-generated subject categories. As users' queries are short and new queries appear all the time, our problem is to assign effective concept categories to users' queries and improve their search performance, even if the given queries have never appeared before. The proposed approach is a well-integrated set of innovative techniques, including relevant concept finding, query clustering, and query classification. A meta search engine, LiveImage, based on the proposed approach is implemented and tested through extensive experiments. The experiment results show that the performance of Web image retrieval can be effectively improved with the proposed approach.