以物體為基礎的影片壓縮技術之研究

Abstract

以物體為基礎的影片壓縮方法是現在很熱門的題目，也是一個影片壓縮很重要的發展方向。藉由將物體由背景中擷取出來，並且針對不同的物體作個別的處理，使得影片壓縮更有效率和彈性。在這篇論文中，我們首先提出一個將剛性(不形變)物體由背景中切割出來的方法，這個方法主要是藉由分析兩張連續影像內容的移動資訊來達成；接著我們提出一個以物體為基礎的影片壓縮方法，利用前一個方法先切割出影片中的物體，我們只需要紀錄物體和背景在之後的畫面中如何移動，以及紀錄一些因為攝影機移動或物體移動而新出現的背景區域像素即可，如此一來，就可以相當有效率地壓縮影片。

Object-based video compression is a hot topic. This kind of techniques separates objects from the background and encodes objects individually. In this thesis, we will propose an object segmentation method and an object-based video compression method. Both methods are fully automatic. The object segmentation approach provides the ability to extract rigid objects from two consecutive frames with or without background motion. The main idea is to analyze the motion information between two frames. Then the result of object segmentation could be used in the object-based video compression method. The object in each frame could be considered the same as the first extracted one. So we only need to record the motions of background and the object in each frame. And the new appeared background area also needs to be recorded. Thus, only a little information is needed to reconstruct the original video.