Abstract: Image re-ranking, as an efficient means to progress the results of web-based image search, has been implemented by p, resent commercial search engines such as Bing and Google. Given a query keyword, a pool of images is initially recovered based on textual information. By inquiring the user to choose a query image from the pool, the remaining images are re-ranked based on their visual matches with the query image. A chief dispute is that the similarities of visual features do not well associate with images’ semantic meanings which infer users’ search intention. Currently people proposed to match images in a semantic space which utilized attributes or reference classes directly connected to the semantic meanings of images as basis. Though, learning a general visual semantic space to typify highly various images from the web is tricky and ineffective. In this paper, we propose an original image re-ranking framework, which routinely offline learns dissimilar semantic spaces for diverse query keywords. The visual features of images are planned into their associated semantic spaces to obtain semantic signatures. At the online stage, images are re-ranked by comparing their semantic signatures attained from the semantic space denoted by the query keyword. The proposed query-specific semantic signatures considerably develop both the accuracy and efficiency of image re-ranking. The novel visual features of thousands of dimensions can be planned to the semantic signatures as short as 25 dimensions. Experimental results demonstrate that 25-40 percent comparative improvement has been attained on re-ranking exactness evaluated with the up to date methods.
Keywords: Image search; image re-ranking; semantic space; semantic signature; Neural Network; keyword expansion.
Title: Re-Ranking Web Images Based On Query-Specific Semantic Signatures
Author: Navamani.C, Anantharaj, Chendavarayan.R
International Journal of Computer Science and Information Technology Research
ISSN 2348-1196 (print), ISSN 2348-120X (online)
Research Publish Journals