Integrated shot boundary detection using object-based technique

Shot boundary detection is a technique used in the first stage of video parsing, whereby shots are extracted before objects are segmented and indexed. Most traditional techniques detect transitions by testing for the feature differences between frames, and tend to fail under long transition with large disturbance. This problem is solved in this paper by considering shot boundary from a different perspective and treating transitions as periods with temporal object discontinuity. A novel concept known as object-based shot boundary detection, which utilises frame lifetime for detection, is introduced. This technique is superior to traditional techniques as it is transition length independent disturbance resistant, transition type independent and does not need a trial-and-error threshold. Object-based detection logically becomes a preferable technique to be used in fully automatic shot extraction systems for the upcoming MPEG-7 platform.