Summary: | 碩士 === 國立臺灣大學 === 資訊工程學研究所 === 92 === With the amazing growth in the amount of multimedia documents, people have become enthusiastic to acquire a more concise and informative representation of these documents. One of the desired technologies is the region-of-interest (ROI) determination. Conventional ROI analysis concentrates on two fundamental types of multimedia documents: image and video. However, the research performance of videos is far behind that of images. The phenomena are arisen from unsuitably considering the essential differences between image and video, and some video’s specific characteristics are ignored.
Facing such a challenging issue, we propose a framework for automatic ROI determination in videos based on user attention model. In this work, a set of attempts on using video attention features and knowledge of applied media aesthetics are made. We classify visual attention features into three fundamental categories: intensity, color, and motion. Referring to aesthetic principles, these features are combined according to the camera motion types on the basis of a proposed video analysis unit, the frame-segment. We conducted lots of experiments on several kinds of video data and demonstrated the effectiveness of the proposed framework. This work is viewed as a preliminary step towards the solution of high-level semantic video analysis.
|