MPEG環境下之時域畫面內差

Abstract

時域畫面內差是一個無最佳解的問題。物體在要插入畫面的時間點上的運 動軌跡通常是不可知的。即使假設了運動模型，要使得插入畫面上的每一 點都有相對應的參照點也是十分困難。在本論文中，我們提出一種利用兩 個參考畫面間移動向量的區塊對應內差方法。這種方法基於平行直線運動 模型的假設來估計對某一參考畫面的初始移動向量，然後再與來自另一參 考畫面的移動向量相比較而調整。這個遞迴調整的步驟一直進行到移動向 量收斂為止。這種方法的優點是它可以輕易的實現在MPEG編碼位元流上。 Frame interpolation in temporal domain is an ill-posed pr- oblem. The object movement trajectory between the time unit when frames are to be interpolated is usually unknown. Even with the assumption of known motion model, it is still diff- icult for all the points in the frame to be interpolated to find the corresponding points in the reference frame. In this thesis, we propose a motion compensated block cor- responding interpolation algorithm for motion vector estima- tion from the two reference frames at the two ends of the t- ime unit. This algorithm is based on the assumption of para- llel linear motion model to estimate the initial motion vec- tor from one reference frame. The motion vector is then adj- usted by comparing the initial motion vector with another m- otion vector from the second reference frame. The adjustment is done with an iterative procedure until the motion vector converges. The advantage of this algorithm is that it can be easily implemented in the MPEG coded bit stream syntax.