New video abstractions for summarisation and browsing

Abstract

In present-day technological developments of multimedia systems, the most challenging conundrum facing the signal processing community is how to facilitate effortless access to a rich profusion of of audio-visual data. In order to enable the intuitive access to large video archives, the areas of video summarisation and abstraction have become momentous. The main challenge of the systems for video summarisation and browsing is to achieve a good balance between removal of redundant sections of video and representative coverage of the video summary. The methods represented in this chapter make a shift towards more user-centred summarisation and browsing of large video collections by augmenting interaction rather than learning the way users create related semantics. A comprehensive survey of the state-of-the-art methods in the area of video summarisation and browsing is presented, followed by a detailed analysis of a novel approach to temporal video representation. In order to create an effortless and intuitive access to the overwhelming extent of visual information, we propose exploitation of universally familiar abstractions such as linking ranking with spatial linear ordering and presented size, projecting highly dimensional data onto the 2D/3D visualisation space as well as using narrative structure of sequential visual art. To achieve this goal, a set of sophisticated yet robust machine learning and knowledge discovery methods are employed. By combining efficient multimedia signal processing and the computational intelligence, coupled with the user-centric interface design, the presented visualisation systems augments the vast abstract space of visual similarity, thus enabling responsive environment for intuitive experience of large-scale video databases. © 2009 Springer-Verlag Berlin Heidelberg.