RE-RANKING WITH CLICK-BASE SIMILARITY AND TYPICALITY USING SPECTRAL CLUSTERING

Abstract: Application on World Wide Web and internet is increasing exponentially. A huge amount of database is added every minute so there is need for effective and efficient image searching. In order to improve image search performance image re-ranking system are proposed, but now days there some problem arises like semantic gap, intent gap which restrict development in image retrieval. This paper present a novel re-ranking approach name spectral clustering re-ranking with clicked base similarity and typicality. We use image click through data as implicit feedback from user and help to overcome intention gap. Proposed system contains saliencing technique. Saliency region contain more important information in the image. So less time require for searching. Our re-ranking approach can significantly improve search results, and outperform several existing re-ranking approaches.

Keywords: spectral clustering, image search, clicked through data, typicality.