基於H.264視訊編碼標準多畫面之快速移動估計演算法

Abstract

移動估計演算法(Motion Estimation Algorithm)在視訊編碼中扮演著重要的角色，經由視訊影像移動估計，可大量減少視訊影像中時間域重複性(TemporalRedundancy)。目前廣受歡迎的國際編碼標準H.264/AVC 為了更進一步的達到高編碼品質及低位元率，採用了多張畫面參考(Multiple reference frames)及可變區塊大小(Variable Block Sizes)的移動估計演演算法。因為H.264 提供這些方法，也因 此增加了計算複雜度，為了降低H.264/AVC 的高複雜度，在本論文中提出了兩個快速移動估計演算法用於提升H.264/AVC的編碼速度，這兩個演算法分別是:(1)根據以移動向量資訊來決定模式(2)根據以區塊模式來決定區塊模式。本論文為了避免暴力法搜尋，而使用了以儲存的移動向量資訊來合成且估計移動中心，本論文藉由分析這些已儲存的移動向量及分割模式，來減少要做估計的模式。透過本論文所提出來的演算法可以有效的提升速度並減少H.264/AVC 移動估計的高複雜度並且維持高品質的視訊品質。

Block-based motion estimation is a widely used technique in the video coding standards to eliminate the temporal redundancy. Through the prediction of object movement between successive frames, only the information of the difference between the continuous frames are encoded for the purpose of the transmission of storage. The video coding standard H.264/AVC provides considerable performance improvement over a wide range of bit rates and video resolutions compared to previous standards. However, these features result in an extraordinary increase in the encoder complexity, mainly regarding to the various coding modes and the multiple reference frames adopted in the ME. In order to reduce the computational cost of the H.264/AVC motion estimation, two fast algorithms Mode Determined by short-term Motion Vectors (MDMV) and Mode Determined by short-term Mode (MDM), are proposed in this thesis. The proposed algorithms MDMV and MDM analyze the short-term motion vectors and short-term coding modes, respectively, to reduce the coding modes used in the ME for long-term reference frames. Experimental results show that, by taking advantages of mode reduction, the proposed algorithms are able to reduce 72% overall ME time while still maintain good performance regarding to bit rates and picture quality.