Detection of Gradual Transitions through Temporal Slice Analysis.pdf

Detection of Gradual Transitions through Temporal Slice AnalysisC. W. Ngo, T. C. Pong R. T. ChinDepartment of Computer ScienceThe Hong Kong University of Science TechnologyClear Water Bay, Kowloon, Hong KongEmail: fcwngo, tcpong, rolandg@cs.ust.hkAbstractIn this paper, we present approaches for detectingcamera cuts, wipes and dissolves based on the analysisof spatio-temporal slices obtained from videos. Theseslices are composed of spatially and temporally coher-ent regions which can be perceived as shots. In the pro-posed methods, camera breaks are located by perform-ing color-texture segmentation and statistical analysison these video slices. In addition to detecting camerabreaks, our methods can classify the detected breaks ascamera cuts, wipes and dissolves in an ecient man-ner.1 IntroductionA video is physically formed by shots; a shot isan uninterrupted segment of screen time, space andgraphical congurations. The boundary between twoshots is called camera break. There are three majortypes of camera breaks: cut, wipe and dissolve. Acamera cut is an instantaneous change from one shotto another; a wipe is a moving boundary line cross-ing the screen such that one shot gradually replacesanother; a dissolve superimposes two shots where oneshot gradually lighten while the other fade out slowly.Wipe and dissolve are normally referred to as gradualtransitions.In the current literature, there are various algo-rithms for detecting camera breaks [7, 10], in general,we can categorize them as statistic-based, histogram-based, feature-based, transformed-based, and motion-based. Most existing algorithms can segment a videosequence into shots correctly if the sequence hassmooth frame transitions within a shot and abruptspatial changes between shots. These algorithms areinsensitive to gradual transitions since the change be-tween two consecutive frames is small. Furthermore,methods based on the frame-to-frame dierence met-rics and feature analysis are computationally expen-