料理映像の特徴を利用した要約手法の検討
We propose a method to abstract cooking videos. We define cooking video abstraction as making videos
sufficient to understand general cooking procedures visually and intuitively. To make an abstracted video, important
subshot segments need to be extracted from the original video. Important segments in a cooking video are cooking
motions and appearances of foods, since visual information that represents essential cooking operation is exceptionally important. These segments have typical motion-related features. Therefore, the proposed method refers to
motion in the image detected as an optical flow. An experiment to detect important segments from motion-based
structural analysis showed high detection rate and promising results. We assembled the important segments, and
the resultant abstracted video showed the effectiveness of the proposed method