以建構背景擷取運動物件之方法

Abstract

以物件為基礎的影片壓縮方法當中，壓縮的效率跟運動物件的擷取是成正比的，若是運動物件的擷取越準確，則其壓縮的效率越好。在此篇論文當中，我們分別對時間域跟空間域做分割，再將兩種結果結合。在空間上我們利用分水嶺演算法，將空間上的特徵資訊找出來。在時間方面，我們將連續畫面相減，然後以自動產生的門檻值判別影像是否有改變，利用此項資訊，建構一個可靠的背景，當一個像素連續多個畫面被判定為沒有改變的像素，則此像素是背景像素，我們便將它建構成背景。最後，將目前的畫面跟背景影像比較而定義出移動物件遮罩。此遮罩再與分水嶺演算法的結果整合即可準確的將物件切割出來。

In the method of the object-base compression video, the compression performance and the motion object segmented are in the direct proportion. If the motion object segmented is more accurate, the compression performance is better. In this paper, we segment the object on the temporal domain and the spatial domain , and then we combine the results of them. In spatial domain, we use the watershed algorithm to find out the characteristics of the spatial domain. In the temporal domain, we subtract the constant frames, and then we determine the frames by the threshold of auto generated if change or not. We make use of this information to construct a reliable background. If the constant frames of pixel are categorized to an unchangeable pixel, the pixel is background pixel. We will construct it to the background. Finally, we compare the current frame with the background to define the motion object mask. The combination of the object mask and the result of Watershed Algorithms is able to segment the objects exactly.