The problem of digital video segmentation occurs in many different applicat
ions. It is the first step toward automatic video indexing, automatic video
editing, perceptual coding, etc. We developed a hierarchical multiresoluti
on shot transition detection scheme which takes advantage of the production
aspect of video for high precision shot transition detection. New features
are proposed to capture the changing statistics of different kinds of shot
transitions so as to identify, not only abrupt shot transitions, but also
gradual transitions accordingly. Identifying complex transitions-dissolve a
nd wipes-is important due to their use as visual cues to story structure in
narrative film. In this paper, a two-pass algorithm, collection and verifi
cation, is proposed to study dissolve and wipe transition detection problem
s, the most difficult shot transition detection problems. Wavelet transform
ation is utilized for image multiresolution decomposition. Experiment resul
ts show that our methodology has much higher precision compared to conventi
onal algorithms. (C) 1999 Academic Press.