低速傳輸中TMN8巨集區塊層量化係數之調整

Abstract

低速影像對於位元分配適當與否十分敏感，本論文研究之重點在於如何盡量減少不必要的位元浪費，去把限定的位元數分配給攜帶許多影像資訊的巨集區塊(MB, macroblock)上。而要真正落實把位元分配到各個MB上，就在於如何調整每個MB所用的量化係數QP。QP變動直接受係數K和C所影響，我們發現TMN8中比重平均控制K和C的方法會造成變動較大之QP，和較多的未攜帶明度(Luminance)和彩度(Chrominance)資訊之MB數目。本論文研究TMN8之速率控制方法後，提出四種以簡單線性回歸(SLR, Simple Linear Regression)為基礎之調整方法來控制K和C。結果發現SLR I和SLR III求得之QP變化較為平穩，所需之標頭位元數(header bit)減少，並能提高攜帶明度彩度資訊的MB數目比例。

The bit allocation efficiency becomes increasingly significant in low bit rate video. In this work, we focus mainly on how to reduce header bit redundancy and allot bits to macroblocks that have rich luminance and chrominance information. It is achieved by adjusting the quantization parameter QP for every macroblocks. The variation of QP is highly related to parameter K and C. We find that the QP estimated from weighted averaged K and C will result in larger variation and more empty macroblocks. After studying the rate control method of TMN8, we proposed four estimating methods based on simple linear regression (SLR) schemes to adjust K and C. The results show that SLR I and SLR III can achieve less QP variation and less empty macroblocks.