New Approach for Re-Ranking of Images

In today's world there is an increasing demand of image search so there is a need for an efficient image search engine. So image retrieval is an important task and moreover images retrieved should map with the query the user has typed. Efficient retrieval of images matching with query must be given prime importance. Most of the search engines use text based search for retrieval of images and few use visual search. If text search and visual image content features are combined together for the retrieval of images it will fetch accurate images and enhances user satisfaction. To map user's intention to the image, a re-ranking framework is used that learns different visual semantic spaces for different query keywords through keyword expansions. The images are retrieved based on query keywords, the visual features of the retrieved images are then extracted. Images are then re-ranked by comparing their semantic signatures obtained from the visual semantic space specified by the query keyword. Thereby, improving the efficiency of image search.