在視訊設備普及的今日，視訊內容的分析與處理工作變得愈來愈重要。視訊內容物件分割方法在視訊內容自動分析應用中是很重要的一項技術。 本論文提出一個快速的視訊內容物件分割方法，所提方法利用去掉均值的影像可保留畫面紋理資訊及降低光源變化與陰影對背景影響的特性來分割物件，可達到快速有效分割物件的目的。實驗結果顯示，我們所提的物件分割方法錯誤率比現有快速物件分割方法低，執行速度則可提升約25%~86%。 In this thesis, we present a novel video object segmentation approach. The proposed approach extracts objects from a frame in a video stream using the difference information between the mean-removed versions of the current and referenced frames. Due to the mean-removed version of a frame reduces the influence of light variation on the frame and reserves the texture information of the frame, the proposed approach can effectively segment objects for video sequences and remove shadow pixels. Experimental results show that the proposed approach has the least computation time among object segmentation approaches with shadow removal capability. Compared with the available approaches, our approach reduces the computation time by 25% to 86% with better segmentation accuracy.
