An object-based shot boundary detection using edge tracing and tracking

With the upcoming development of MPEG-7, research in shot boundary detectio n has become more popular. Traditionally, shot boundary has been detected b y measuring the differences between the features of adjacent frames, and th e threshold was constructed based on a trial and error basis, which usually achieved an accuracy of nearly 90%. However, the remaining errors were lef t unresolved. In this paper, we will tackle the problem by using object-bas ed detection, which utilizes information from multiple frames through a tim e stamp transferring mechanism. Unlike classical techniques, our method can handle a gradual transition of any length. The concept of edge object trac king is developed to track moving objects as well as unbounded objects acro ss the gradual transition scenes. We will also prove that object-based dete ction is not merely an algorithm that can detect shot boundary, but is supe rior to traditional indicators for detecting gradual transition. Such a tec hnique works very well with frames containing skipped blocks and is indepen dent of the transition period and the amount of disturbances within the seq uence, which cannot be handled by indicators based on adjacent frame compar ison. (C) 2001 Academic Press.