Image re-ranking is a very effective technique to improve the results of web based image search. Most of the current commercial search engines (Bing, Google) are adopted this technique. In this paper new technique is proposed to improve the results of web based image search. In this technique, at the offline stage, different semantic spaces for a different query keyword are learned automatically. After that semantic signatures are obtained by projecting visual features of the images in to their related semantic spaces and hashing technique is used to compact obtained semantic signatures. At the online stage, compacted semantic signatures of the images are compared to re-rank images. This new technique improves accuracy and precision of web image search.
Loading....