QQ客服 官方微博 反馈留言

	Query-Specific Visual Semantic Spaces for Web Image Re-ranking
	Xiaogang Wang; Ke Liu; Xiaoou Tang
	2011
会议名称	IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
会议地点	Colorado Springs, CO
英文摘要	Image re-ranking, as an effective way to improve the results of web-based image search, has been adopted by current commercial search engines. Given aquery keyword, a pool of images are first retrieved by the search engine based on textual information. By asking the user to select a query image from the pool, the remaining images are re-ranked based on their visual similarities with the query image. A major challenge is that the similarities of visual features do not well correlate with images' semantic meanings which interpret users' search intention. On the other hand, learning a universal visual semantic space to characterize highly diverse images from the web is difficult and inefficient. In this paper, we propose a novel image re-ranking framework, which automatically offline learns different visual semantic spaces for different query keywords through keyword expansions. The visual features of images are projected into their related visual semantic spaces to get semantic signatures. At the online stage, images are re-ranked by comparing their semantic signatures obtained from the visual semantic space specified by the query keyword. The new approach significantly improves both the accuracy and efficiency of image re-ranking. The original visual features of thousands of dimensions can be projected to the semantic signatures as short as 2 5 dimensions. Experimental results show that 2 0 % 3 5 % relative improvement has been achieved on re-rankingprecisions compared with the state-of-the-art methods.
收录类别	EI
语种	英语
内容类型	会议论文
源URL	[http://ir.siat.ac.cn:8080/handle/172644/3268]
专题	深圳先进技术研究院_集成所
作者单位	2011
推荐引用方式 GB/T 7714	Xiaogang Wang,Ke Liu,Xiaoou Tang. Query-Specific Visual Semantic Spaces for Web Image Re-ranking[C]. 见:IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Colorado Springs, CO.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。