適用於畫面間小波轉換視訊編碼之改良式移動補償時間濾波技術

Abstract

在本項研究中，我們將改良式移動補償時間濾波技術(motion compensated temporal filtering)整合進畫面間小波轉換視訊編碼(interframe wavelet video coding)之架構中。畫面間小波轉換視訊編碼可同時提供畫質、時間解析、與空間解析之可調性。此外，在高位元率時，它具有與H.264/AVC可相比較之編碼效率。在本論文中，我們針對在移動補償時間濾波與移動資訊可調性技術進行探討。 移動補償時間濾波是一個在決定編碼的表現非常重要的要素。我們修改源自於H.264/AVC之移動估測方法與其編碼方式，並將其運用至畫面間小波轉換視訊編碼架構中。並且，我們也將框內方塊(I-block)、雙方向移動估測(Bi-directional motion estimation)、及移動成本函式調整等技術運用至其中。此外，我們針對以h.264/AVC為基礎之畫面間小波轉換視訊編碼提出移動資訊切割技術(motion information partitioning)。藉由針對移動向量進行不同精確度與搜尋方塊大小的編碼，我們可以將移動相量切割為多層次(multi-layer)。根據可調性的要求，如空間解析、位元率或複雜度考量等，可決定解碼所需之層次數目。由實驗結果可發現其在主觀視覺感受的改良

An enhanced motion compensated temporal filtering (MCTF) coding scheme is incorporated into the interframe wavelet coding architecture in this study. Interframe wavelet video coding can provide SNR, temporal, and spatial scalability simultaneously. Furthermore, it has a comparable coding performance to H.264/AVC at high bit rates. In this thesis, we propose improvements on MCTF and motion information scalability. Motion compensated temporal filtering (MCTF) is an essential component in deciding the coding performance of an interframe wavelet system. We modify the motion estimation syntax/scheme originally specified in the H.264/AVC and fit it into the interframe wavelet structure. In addition, the techniques of I-block, bi-directional motion estimation, and motion cost function adjustment are employed to improve image quality. Furthermore, in the original interframe wavelet video coding, the motion information is non-scalable and it may become the dominate portion of the entire bitstream at low bit rates. We propose a motion information partitioning scheme applied to the AVC interframe prediction. By encoding motion vectors at different accuracy and block-size level, we split motion vectors into multiple layers. Experimental results show very promising performance improvement particularly on the subjective quality.