Motion Based Automatic Abstraction of Cooking Videos

In this paper, we propose a method to abstract cooking videos. We define cooking video abstraction as shrinking videos maintaining the understandability of general cooking procedures visually and intuitively. Cooking motions and appearances of foods are considered as important segments in a cooking video. A method to extract such important segments referring to the intensity of motion in an image is proposed, and effectiveness of the method is shown through evaluation experiments. We also developed a cooking video abstraction system that assembles important segments detected by the proposed method and repetitious motions that is especially important among cooking motions. The resultant abstracted videos were about 1/8 to 1/12 of the original videos in time, maintaining the understandability. The validity of the abstraction method was checked by comparing the abstracted videos with manually abstracted videos provided from a broadcasting station.