抄録
Technologies to segment videos into “scene” units consisting of the shot sequence captured in a same place have been useful for program video retrieval and video summarization. We proposed scene segmentation methods based on the change of color formation produced from color histogram , and the change of block-image-kinds formation considering not only color but also textural properties. However, these methods have problems that slight color noises and drastic changes of camera angle could cause false positives. In this paper, we report on our new proposal method of scene segmentation adopting the color formation based on “superpixel” as an unit of image segmentation to solve problems in our conventional methods.
