密集真實移動場估測於視訊壓縮之應用

Abstract

本論文針對H.264/AVC標準中的B-畫面，提出一種新的編碼方法。此方法亦是使用畫面間動作補償來預測影像，並且傳送剩餘資訊來校正影像；唯不同的是，影像的預測方法是在解碼端重新估算密集真實移動場，並以之內插影像，這當中編碼端不須傳送任何動作資訊給解碼端。我們將此方法以 16x16、8x8 與 4x4 這三種方塊大小整合入 JM17.2，並且為其帶來編碼優勢。結果也顯示，此方法在視訊壓縮中仍大有可為。

In this thesis, we propose a new coding mode in H.264/AVC B-frame. The new mode consists of two parts: predictor and residual part. The predictor is generated by motion-compensated frame interpolation (MCFI), and the residual part is calculated in encoder-side and is transmitted. The MCFI algorithm is formed by state-of-the-art dense true motion estimation and dense motion compensation (DMC). The predicting procedure needs no information in decoder-side, so the entropy is cut down to the minimum. The proposed new mode is integrated into H.264 JM17.2, and has three different block sizes: 16x16, 8x8, and 4x4. Since the new mode predictor achieves quite well qualities than H.264/AVC in good amount of region, and the proposed new mode takes significant percentages in several sequences, the new proposed mode is quite promising in video coding and worth further studying.