Contents Scene Extraction Based on Distance Matrix

This paper shows the solution for automatic scene extraction from video footages. The purpose of this paper is development of automatic and semi-automatic video editing technology. This scene extraction algorithm has three staged cascade process. The first stage is scene analytic feature extraction using content deep network encoding information. The second stage is using temporal network for video feature extraction. And the last stage is clip video feature generation by maximum activation of convolution and normalization. A deep learning network is adapted for mixture with scene analysis feature and video feature information. This module showed 0.57 F1 score.