"Extraction of Individuals from Talk Show Programmes Using Shot Detection and Unsupervised Clustering"

Supervised by Dr JP Barker

Abstract

At a time when access is available to innumerable hours of video data, it is extremely important to end users to be able to quickly scan through large quantities of video data in a very short time. Video summarisation techniques endeavour to extract the most important and relevant details into a compressed form so as to provide quick and easy navigation.

This project delivers a tool which, when presented with an episode of a 'talk-show' television programme, outputs a series of still images that ideally depict the host and all guests who feature in the episode, to a recognisable standard. The system performs significantly better than summarisation approaches which use temporally evenly spaced images, and whilst optimal proficiency is not attained, the system may be used for practical applications. The findings of this project are additionally very beneficial to guide future work in the field of automatic video summarisation.