Browsing and searching MPEG-7 images using formal concept analysis

• Authors: Jon Ducrou;Peter Eklund
• Venue: AIA'06 Proceedings of the 24th IASTED international conference on Artificial intelligence and applications
• Year: 2006

Content-based search and browsing of image collections is a difficult problem. This paper argues that Formal Concept Analysis (FCA) is ideally suited to this task and engineers a solution to demonstrate the idea. In general, standard keyword search is problematic when dealing with images. First, the value and names of the metadata are often unknown. Second, image search is about similarity in the sense of combinations of location, shape, luminance etc. These properties are hard to evaluate and unsuited to vector-space models of similarity used in information retrieval. This paper presents a Web-based FCA system for browsing MPEG-7 images and user evaluation results. The system is called IMAGE-SLEUTH. The trial collection that we use to profile IMAGE-SLEUTH are image objects based on a collection of items from the popular game The Sims 2. User testing was performed on 29 university students from various disiplines. The Sims data set is ideal for two reasons. First, The Sims 2 imposes an interesting information space over the items which are annotated with scales such as needs and function, as well as the capacity to enhance the Sim character's skill: objects which allow access to other skills such as logic, creativity, charisma. This information space is revealed by the concept lattice. Second, the Sims items can be used to demonstrate the value of search via image descriptors such as shape, colour and luminance. MPEG-7 is therefore an important platform for image annotation since it provides for the information space induced by the game playing attributes as well as visual descriptors. This paper describes the process of image extraction and a description of the Web-based image browser - IMAGE-SLEUTH.