Date on Master's Thesis/Doctoral Dissertation
Computer Engineering and Computer Science
Image searches; Image retrieval; Data mining; Focused crawling; Data extraction
Images, Photographic--Databases; Indexing--Computer programs
Recently, social multimedia-sharing websites, which allow users to upload, annotate, and share online photo or video collections, have become increasingly popular. The user tags or annotations constitute the new multimedia "meta-data". We present an image search system that exploits both image textual and visual information. First, we use focused crawling and DOM Tree based web data extraction methods to extract image textual features from social networking image collections. Second, we propose the concept of visual words to handle the image's visual content for fast indexing and searching. We also develop several user friendly search options to allow users to query the index using words and image feature descriptions (visual words). The developed image search system tries to bridge the gap between the scalable industrial image search engines, which are based on keyword search, and the slower content based image retrieval systems developed mostly in the academic field and designed to search based on image content only. We have implemented a working prototype by crawling and indexing over 16,056 images from flickr.com, one of the most popular image sharing websites. Our experimental results on a working prototype confirm the efficiency and effectiveness of the methods, that we proposed.
Zhang, Zhiyong, "Focused image search in the social Web." (2008). Electronic Theses and Dissertations. Paper 1639.