부산대학교 도서관

Detection of interesting part from video on web or in large database with less time consumption is a big issue. Decomposition of video into shots and scenes with key frame extractions are useful to some extent with the compromise in spatio-temporal characteristics. We have presented techniques for frame base shot boundary detection, scene change detection and key frame extraction in videos, which are useful to differentiate in presence of any type of motions. It is useful to define particular frame ranges of gradual transitions as a fade and dissolve as sequence of frames in presence of motion, which are difficult to detect compare to the sharp transitions. Approach is based on the discrete cosine transform based DC Coefficients and DC image, where mean and variance of DC image values are sufficient to choose threshold to detect these transitions effectively. This concept was applied and verified on large database; and here we presented results applied on different movies. Performance is verified with existing techniques in terms of recall, precision measures and results reflect more about this concept at the end.