Recommendation Method to Make Combined Video from Video Segments

摘要

In this paper, we propose a recommendation scheme for media framework, which enables users to make their own videos by writing a story and reusing parts of accumulated videos. To reuse part of videos, we split a video into semantic segments based on an analysis of relations among objects in a video and store the segments with semantics in repository. To create a new video, the user sends queries based on his/her own story, then the framework recommends appropriate video segments for each query. To determine the rank of searched segments, the recommendation engine uses the degree of coincidence between the segment and the query. Also, it uses the degree of similarity between the searched segment and previously selected segment. By doing this, we can recommend a segment, which is consistent with user's intent and harmonized with the other parts of the new video.