A NOVEL APPROACH TO KEY-FRAME DETECTION IN VIDEO

摘要

Key-frame selection is the key to effective cataloging and browsing video, In this paper, we present a novel method that adopts coarse-to-fine strategy to detect structure and color changes to select key-frames. These measures not only filter negative false alarms caused by illumination changes but also improve recognition rate of the keyframes. Our results show that better results are obtained with the use of two similarity measures that we proposed in addition to the Earth Mover's Distance, as opposed to using Earth Mover's Distance alone.