新聞影片視訊場景切割及摘要之研究

Abstract

本篇論文是研究在電視新聞檢索系統中有關場景切割以及視訊摘要的部份。在偵測場景切割 的部份，我們先利用計算直方圖距離(histogram distance)和像素點距離(pixel distance) 來選擇場景切割的候選點，再利用區域最大值(local maximum)的測試將非場景切割的點(false positive)去除。 在視訊摘要的部份，則分別以靜態和動態兩種方式來完成。我們以找出字幕，閃光燈，人物特寫 鏡頭等含有重要資訊的影像(keyframe)來作為靜態的視訊摘要。而以將高動量，靜態，含閃光燈 以及含人物特寫鏡頭等重要場景綜合並加以適當的調整後，來完成動態的視訊摘要。

In order to accomplish the segmentation and video summary in our News Video Browser System(NVBS), in this paper, we propose a robust scene change detection algorithm and two different video summary method, static summary and dynamic summary. In detect scene change point,we compute histogram distance and pixel distance for each frame pair to select possible scene change point. After this, we use local maximum test to eliminate the false positive. For each shot, we find the caption picture, flashlight picture and close-up picture to be keyframes.In dynamic summary, we select the high motion shot, static shot, flashlight shot and close-up shot. After some adjustment, we carry out the dynamic summary.