本研究係以主觀視訊品質評估為基礎,探討視訊摘要(Video Abstraction)技術中的動態影像視訊摘錄(Video Skim),以三種不同的擷取序列片段長度(Skimmed Segment),分別以2.5秒、5秒與10秒,作為擷取序列片段的固定單位長度,以及三種不同的視訊摘錄總長度(Total Skimmed Length),分別以原始影像檔的5%、10%與15%,觀察受測者對上述不同組合,所代表原始視訊內容的差異程度。 我們應用媒體豐富理論中媒介個別化焦點(Personality Focus of the Medium)與多重線索(Multiple Cues)因素,來提高動態影像視訊摘錄(Video Skim)於劇情片電影(Feature Film)的摘錄代表性。 我們發現較長的擷取序列片段長度與視訊摘錄總長度之組合有較佳的代表性,若再以集中結局的擷取與包含文字、音訊及視訊的擷取策略,可使得摘錄視訊更具有原始影像之代表性。
Video skim combined total skimmed length(TSL) with skimmed segment(SS) based on a subjective quality assessment method is examined in this paper. In our experiment, the total skimmed length includes three levels, 5%, 10%, and 15%, while the skimmed segment includes three levels, 2.5, 5, and 10 seconds for a separated unit. We observe that the video skim with the longer TSL and SS has the better representativeness to original movie content, but the four combinations for 10% and 15% with 5 sec and 10 sec are insignificantly different. The finding is helpful for reducing the time cost of skimming video. Further, we apply two important factors, personality focus of the medium and multiple cues, from the media richness theory to our skimming method in order to raise the representativeness of video skim for a feature movie. In the personality focus of the medium, we identity three parts for a feature movie and discuss three different combinations to focus on each part averagely, many ending part, and most ending part. We find that skimming the most end part and including multiple cues such as text, audio, and video could raise the representativeness of film skim.