針對畫質與頻寬限制的串流系統，使用自適性跳畫面機制的初步探討

Abstract

由於人眼對於畫質好壞差異的感受比斷斷續續畫面所帶來的停頓效果較不 敏感，大多編碼器都會注重在將可用位元數分配到所有可編畫面上，即使個別畫 面會因此得到較少的可用位元而使得畫質變差，也大多不願意將畫面率降低來得 到較好的畫質。但是對於監控系統或是一些比較在意畫面品質的攜帶型設備而 言，則會對於編碼畫面品質上有一定要求。 因此如何在同時具有畫質與頻寬限制的環境達到最佳化變成我們想要解決 的議題。為了解決多重限制的問題，我們首先從了解最佳解的設計出發，再深入 研究如何輔以Lagrange multiplier 的方式來達到位元率- 失真交換下 (Rate-Distortion Trade-Off)最佳化。 相較於一般流量控制的設計都是只考量頻寬限制的要求，我們的問題勢必 要使用動態規劃才能達到最佳位元分配；然而為了降低複雜度，支配線 (Dominative Line)與依次精修(Successive Refinement)的方法被提出並且分析 其結果。

Owing to that human eyes are more sensitive to jerky effect caused by different frame rate rather than quality variance caused by different quality of coded frames, most encoder systems focus on distributing available bit budget among all frames and are not willing to reduce frame rate to obtain better spatial resolution, even the quality of each frame becomes worse for getting less available bits. However, there exists quality requirements to surveillance systems and some mobile devices which care about the quality of each coded frame. As a result, how to solve the problem with both distortion- and rate-constraint is the main issue in our research. In order to solve this multiple constrained problem, we start by understanding the design of optimal solution; furthermore, we study how to use Lagrangian parameters for rate-distortion trade-off optimization. Compared with general rate control scheme, which only consider the rate/budget requirement, our problem must use dynamic programming for optimal bit allocation. Nevertheless, in order to reduce complexity, dominative line and successive refinement methodology are proposed and analyzed.