位元率-誤差最佳化量化器設計

Abstract

位元率-誤差最佳化量化能增進影片壓縮效率。然而，目前已提出之基於操作性的方法因需要進行窮舉搜尋，往往導致偏高的運算量。本篇論文中提出一個新的基於分析性之位元率-誤差最佳化量化方法。利用一個熵編碼之位元率模型，我們提出了最佳量化方法的解析解。與H.264/AVC的參考軟體相比，在使用IBBP模式編碼於常用PSNR範圍(30-40dB)時，此方法在相同畫質下減少了約5-7%的位元率，同時只增加了最大不超過1%的總編碼運算時間。

Rate-distortion optimized quantization improves the coding performance of video compression. However, most existing methods based on the operational approach require an exhaustive search process that is computationally expensive. In this thesis, we present a novel analytical method that provides a closed-form solution using a rate model of entropy coding. The computational cost of rate-distortion optimized quantization is greatly reduced. Our method achieves 5-7% bitrate reduction with no greater than 1% increase of total encoding time over the H.264/AVC reference encoder for video coded with IBBP GOP structure in normal PSNR range (30-40dB).