Digital Multimedia objects, in libraries or otherwise, hold the promise of being able to be indexed by their contents just as other documents can be indexed and searched by their full text. This talk presents new paradigms for digital knowledge management that were developed in my lab; they include

- novel browsing methods that utilise an automated structuring of the digital collection as a small-world graph with a low degree of separation

- content-based multimedia retrieval, eg, search of still images similar to given ones; finding a news story broadcast over TV by providing visual examples; finding a music piece by humming it etc. Our approach for finding material is not based on manual annotation but on /automated/ processing.

- clustering and visualisation techniques to present material in a collection or organise large document sets that were returned in a query.

Some of the challenges in this approach are given by the semantic gap between what computers can index and high-level human concepts and by polysemy, ie, the many meanings and interpretations that are inherent in visual material and the corresponding wide range of information need by the user. We try to overcome these challenges by utilising the skills of the user, for example through a process that we call relevance feedback, thus putting the user at centre stage. Behind the scene, we deploy learning algorithms that adapt themselves to the user and their information need.

We argue that these methods, when integrated into digital libraries, will not only enhance their searching and browsing capabilities but also give access through unconventional query methods such as sketching, similarity browsing and providing examples of what is relevant.

Download PowerPoint presentation (10.4Mb ZIP file)

Watch the webcast replay >>