在傳輸緩衝器限制下影像之最佳化延遲編碼理論

Abstract

近來完成的影像壓縮標準，如CCITT 的H.261 和ISO 的 MPEG1，是建立在 使用移動補償 (motion compensation)、空間領域的餘弦轉換 (DCT)、純 量量化 (scalar quantization)、霍夫曼編碼(Huffman coding) 等所構 成的混合編碼技術之上。這些標準詳細訂定了壓縮編碼器之壓縮資料的傳 輸格式以及接收端的解壓縮方法，卻也保留了相當大的設計寬容度。根據 這自由性，我們提出一個稱為”延遲編碼〞的理論，來克服一些典型編碼 機構所產生不適宜的人為偏差。最顯見的例子為緩衝區的溢滿 (overflow) 與空乏(underflow) 以及不均勻的影像品質分佈。這些現象 產生的原因是由於傳統編碼器中簡易地使用緩衝器充滿度(buffer level) 來回授控制量化器的步階(stepsize)尺度。關於這計畫的理論發展可劃分 為二部分：一、為一般性延遲編碼方法的研究以及它的最佳化併考慮緩衝 器容量限制。二、為根據數種不同的均方誤差(MSE) 準則以發展一套特定 的演算法來完成最佳延遲編碼。根據這架構，實驗結果顯示出：壓縮影像 的品質較傳統的標準方法佳，而且緩衝器的充滿度能被完善的預測及控制 。此一最佳延遲編碼的基本理論可應用在 H.261及 MPEG1之外的一般編碼 演算法中。 The recently completed CCITT H.261 and ISO MPEG1 video compression standards are based on a hybrid coding technique which employs motion compensation, spatial-domain discrete cosine transform, scalar quantization, and Huffman-type coding. They both specify the syntax of the compressed data stream and the method of decoding, but leave considerable latitude in the design of the encoder. We therefore propose a coding approach, called delayed coding, to overcome some undesirable artifacts which may accompany typical coding schemes based on simple buffer-level-feedback control of quantization stepsizes. Example of such artifacts are buffer over- and underflows and non-uniform visual quality across different parts of an image. Our theoretical studies on this approach can be divided into two broad parts, i.e., (1) the investigation of a general delayed-coding approach and its constrained optimization and (2) the development of a particular algorithm for optimal delayed coding within the H.261 framework based on several different MSE criteria. Experimental results show that, with this approach, the compressed video quality can be improved over typical coding methods. Also, the buffer level can be perfectly controlled. The basic approach of optimal delayed coding is applicable to coding algorithms other than H.261 and MPEG1.