This project is based on image search with re-ranking in the web-based by the given query keyword. Normally people used the common search engines for retrieve the image such as google, bing. In that time search engines displayed the large no of images. Users select the particular image from the pool of images. One of the challenges is that the visual features have a lot of similarities and hence do not correlate well. Now a days, people searching the images by image’s attributes or reference classes. But different or highly diverse images searching from the web is tuff and poses a lot of inefficiency.

Here project proposed a novel image re-ranking frame work, which the different semantic spaces for different keywords learns automatically. To obtain semantic signatures the featured visuals of images are estimated into their corresponding semantic spaces. In online stage, a comparision of semantic signatures is made and then the images are re-ranked .These semantic signatures are acquired from the semantic space stated by the query keyword. In the postulated query-specific semantic signatures notably enhance the image re-ranking in both accuracy and efficiency.

Keywords: Re-ranking, Semantics, Signatures, Web image