Dissolve Detection Based on Twin-comparison with Curve Fitting

Video shot boundary detection is a fundamental step in video indexing and retrieval. There are a number of different types of transitions or boundaries between shots. Most shot change detection algorithms can detect abrupt transitions without difficulty. However, gradual transitions usually consist of wipes, fades, dissolves and computergenerated effects, so it is much more difficult for us to detect them than to detect abrupt transitions. In most available dissolve detection algorithms, the false/missed hit rate caused by motion is relatively high. In this paper, we propose a novel method customized for dissolve detection. The twin-comparison approach is first utilized to roughly detect dissolve candidates. Then a special curve is fitted according to the inherent characteristic of dissolves indicated by two novel parameters, i.e., the percentage of dissolve-supporters in a frame and the variance of coefficients in the dissolve model. At last, we utilize this curve function to remove non-dissolve transitions. Experiments performed on various types of video clips validate the proposed approach. Compared with the twin-comparison scheme and the linear algorithm, our scheme can greatly improve the precision performance for all kinds of video clips with a bit reduction in the recall performance.