基於位元-失真理論之H.264/AVC位元控制研究

Abstract

位元控制在訊號處理上扮演相當重要地位，對於視訊壓縮的應用更是重要。其中，位元控制有兩個取捨點作為相互影響的因子，一為資料品質，也就是失真度，另一為資料量。在視訊壓縮系統上，共定義出兩種不同的影像壓縮型別，分別是Intra-Frame與Inter-Frame形式。 Intra-Frame扮演整個影像壓縮的起始點，除了要保持高度清晰的影像品質外，在位元的使用上更需精確應用，主因為接下來的影像需參考此一影像並使用剩下的位元進行壓縮。為了增加壓縮效能，或是減低位元需求，常會加入大量的Inter-Frame形式的影像。在即時的視訊壓縮系統中，藉由前一張影像的資料品質與資料量來快速預測量化係數是位元控制所扮演的腳色。然而，真實的影像資料是會隨物體以及時間產生變化，往往兩張畫面就有可能產生極大的落差。因此，如何藉由前一張影像的資料品質與資料量來預測出精確的量化係數，因而產生穩定的影像品質或是使用精確的位元數成為重要的研究課題。 本研究以視訊壓縮H.264/AVC為主，架構於原始的位元-失真理論，而此基準點也是目前H.264/AVC在位元控制所採用的方法，探討其現階段位元控制的法則，得到一個可行性的失真度預測方法，進而將之應用於Intra-Frame與Inter-Frame的兩個層面。更近一步的討論，基於這兩個層面，在不同的應用考量上，共推導出四個不同的位元控制演算法。

基於Intra-Frame的位元控制演算法。 利用所推導得到的失真度預測方法，藉由常見的標準測試影像進行參數訓練，進而得到快速且精確的Intra-Frame位元控制演算法。 基於Inter-Frame的位元控制演算法。 利用所推導得到的失真度預測方法與位元估測方程式，透過Lagrange最佳化的推導，提出一個解析性的精確位元控制演算法，改善原本H.264/AVC以及現有著作在低位元率限制下表現不佳的缺點。 在實務應用上，優先考量嵌入式系統的適用性。藉由H.264/AVC內建的量化係數預測技術，得到初始的量化係數，利用所推導得到的失真度預測方法以及影像直方統計曲線，進行量化參數修正。 藉由所推導出的方程式來有效的預測影像複雜度，以達到較先前研究更精確量化參數數值。目的在於解決低位元率的視訊傳輸之應用。

由原始的位元與失真理論，推導出失真度預測方法。為驗證此一方法其效能的適用性，本文基於四種重要且常見的應用，將此一方法加入，藉由實驗結果指出，此一方法確實能有效運用位元來提升影像壓縮品質。此外亦有將嵌入式系統的硬體限制考慮，讓此方法更能貼近產業界應用。

Rate control plays a vital role in signal processing and video compression. Two degrees of distortion and consumed counts for measurement are selected in a tradeoff consideration. The video compression system defines two distinct frame types, as follows: intra-frame and inter-frame. The start frame in encoded video sequences is assigned as an intra-frame, and the goal is to maintain a higher image quality and lower encoded bits because the intra-frame is referenced after the encoded image, and the remaining bits are used for compression. In a real-time video compression system, referred to as H.264/AVC, the determination of the quantization parameter by rate control must refer to the encoded results in the previous frame. However, sudden occurrences such as fast-moving objects or complex backgrounds in two neighboring frames are usual. Thus, because of encoded results in previous images, producing stable image quality or accurate bits consumption is a crucial topic. This study focused on the H.264/AVC video compression system and rate-distortion theory to obtain a workable distortion prediction method for intra-frame and inter-frame rate control. Subsequently, we used this distortion prediction method to develop four rate control algorithms for various applications.

Based on the intra-frame rate control algorithm According to the distortion prediction method and a large number of test images, we obtained a set of system parameters to develop a fast and accurate intra-frame rate control algorithm.

Based on the inter-frame rate control algorithm We propose an analytic inter-frame rate control algorithm according to the distortion prediction method, H.264/AVC rate prediction method, and Lagrange optimization. This algorithm can solve the problem of low bit rate limitation in H.264/AVC.

In real applications, we first considered the limitation of embedded systems. The initial QP value can be obtained through the built-in technology of QP calculation. We developed the intra-frame rate control algorithm according to the initial QP value, the distortion prediction method, and the information of image histogram.

In the proposed distortion prediction method, we can predict the measurement of image content complexity efficiently to obtain a more accurate QP value than by using the current algorithm. This algorithm was developed for low bit rate video conferencing.

We derived the distortion prediction method according to the original rate-distortion theorem. We applied the method on four differing applications to verify the results. Based on the simulation results, the method improved compressed image quality efficiently. We also considered hardware limitations for industrial application.