Automatic scene/DVD chapter extraction in Hollywoodian movies

Abstract

In this paper we tackle the issue of video scene/DVD chapter segmentation. First, we introduce a review and analysis of the most salient methods existent in the technical literature. Then we propose a novel methodological framework for high level video temporal structuring and segmentation that extracts scenes/DVD chapters based on temporal constraints clustering, adaptive temporal lengths, neutralized shots and adaptive thresholding mechanism. The output of our method provides a structured video and facilitates the user access to different parts of the image sequence. In order to validate the proposed technique we have considered as low level visual features the interest points extracted using the SURF descriptor. The experimental evaluation validates the proposed approach returning an average score for the F1 norm of 82%.