Shot Boundary Detection Based On Image Segmentation

The most fundamental task in video retrieving is to partition long video sequences into a number of shots. In this paper, an overview of some existing shot boundary detection algorithms is given, after that, a shot boundary detection method based on distinct visual patterns to appear on the Spatio-Temporal Slices is proposed to effectively detect both abrupt and gradual shot transitions including cut, dissolve, wipe and fade transitions.An improved cut detection algorithm is proposed. Firstly, the gradient information is computed based on motion compensated Spatio-Temporal Slices, then, adaptive threshold is used to locate cut boundary from discrete value.A new gradual shot transitions detection algorithm based on pixel slope is proposed. To dissolve, the least error slope among three direction of horizontal, diagonal and back-diagonal is used as pixel slope, a 2-D image is build by slope and error, then, this image is transformed to binary image and mathematical morphology filters is used to eliminate the noise. The fade boundary is found using the feature of fade boundary contains monochrome frames, then, the start and end position is detected by dissolve detection algorithm. Wipe is classified into cut-wipe and dissolve-wipe based on the features of the joint of adjacent two shot in wipe boundary. The modified cut detection algorithm is used to detect cut-wipe and the modified dissolve algorithm is used to detect dissolve-wipe. Experiment demonstrates it enable to locate gradual shot transitions exactly and quickly.