Content-based Image Retrieval on the Web Scale (VisualWeb)
- Project Identification
- Project Period
- 1/2010 - 12/2012
- Investor / Pogramme / Project type
- Czech Science Foundation
- MU Faculty or unit
- Faculty of Informatics
- similarity search; content-based retrieval; distributed indexing; scalability
Digital images form a large part of the Web content and proprietary image collections and their amount grows at enormous speed. Current techniques for searching in images are not satisfactory because search engines restrict image search to associated text or attribute annotations. Although content-based image retrieval techniques are improving with respect to effectiveness, they are still limited in the volume of images they can handle - the scalability is still a serious problem.
This project aims at developing an innovative network technology platform which will allow organizing billions of images and searching by visual content. To enhance perceived quality of search results, specific techniques for clustering and ranking of results will also be developed. The platform will be tested on large image collections to demonstrate its properties. Several institutions and companies offering services such as online content-based retrieval, detection of plagiarism, filtering pornographic material, and brand reputation analysis based on logo detection, will cooperate as data providers.