Image Re-ranking is an effective way to improve the results of Web-based image search, has been adopted by search engines Given a query keyword, a pool of images is first retrieved based on the textual information’s. The user has to select a query image from the pool; the remaining images are re-ranked based on the query image. The main challenge is that for different query images, the effective low-level visual features are different. It was difficult to cover large diversity of all web images which makes the query image to classify to a wrong category. In order to reduce the semantic gap, Query-specific semantic signature was introduced. We proposed NOVEL-FRAMEWORK for web image re-ranking, in offline learns different semantic spaces for different keywords individually and automatically. At online stages, images are Re-ranked by comparing their semantic signatures obtained from the semantic space specified by the query keyword. Query specific semantic signatures improve accuracy and efficiency of image re-ranking. Redundant image filtering process is integrated with the system. Query expansion is upgraded using query patterns and associations. The approach significantly improves the precision of top-ranked images and also the user experience.