用於視訊編碼上之數個新運動估測演算法

Abstract

視訊壓縮是視訊傳輸與儲存的必要步驟。運動估測則是視訊壓縮除去時間 上的重複性與達成高壓縮率的核心。而區塊比對法是視訊壓縮標準最常採 用的運動估測方法。本論文分成兩方面提出新的運動估測方法。首先，傳 統的區塊比對法都是以空間上的相似性作為比對標準，此與現有的視訊壓 縮標準採用的轉換域編碼並不相符。基於此觀念，本論文提出新的以離散 餘弦轉換域為比對標準的運動估測法。由於直接運用空間上的運動估測方 法於離散餘弦轉換域將會導致相當高的計算複雜度，本論文提出了新的遞 迴離散餘弦轉換法來解決此問題。本論文提出的另一方面的新運動估測法 是基於 MPEG II 的空間調變模式。此模式提供各種的空間解析度，且是 MPEG II 提供的四種模式之一。本論文基於此模式提出了三種新的區塊比 對法。與現存最佳的全域搜尋區塊比對法比較，此三種新的方法均有更佳 的壓縮率、訊噪比、與計算複雜度。在 MPEG II的基本層運用轉換域的運 動估測法、而在增益層採用本論文提出的三段運動估測法，此時系統與採 用傳統的運動估測方法之系統相較，有相當優秀的表現。 Video compression is essential for video transmission and storage. Motion estimation is a kernel step to eliminate the temporal redundancy and achieve high compression rate. Block- matching algorithms have been widely used for video compression standards. In this thesis, we propose new motion estimation algorithms in two aspects. First, the matching criterion of the traditional block-matching algorithms is in spatial domain, which is not very amenable for the transform coding utilized by contemporary compression standards. From this aspect, we consider new motion estimation algorithm with DCT-domain matching criteria. We show that a direct applying of spatial- domain block-matching algorithm to DCT-domain leads to a tremendously high computation complexity. Hence, we propose a recursive DCT algorithm to solve the problem. Also, we consider a new matching criterion for the DCT-domain. The second aspect of the new algorithms is based on the mode of spatial scalability in MPEG II. Spatial scalability, which is designed for various spatial resolution, is one of the four basic modes in MPEG II. We present three new block matching algorithms for this mode. All these three algorithms have better performance in bit-rate, SNR, and computation complexity compared with the full-search block matching algorithm, which has been considered as the best method in spatial domain. The simulation results show that a three-pass algorithm among the three algorithms provides best performance. Applying the transform-domain motion estimation algorithm in the base layer and the three-pass motion estimation algorithm in the enhancement layer of MPEG II, the system has an outstanding performance compared with applying traditional motion estimation algorithms.