調變式區塊壓縮法

Abstract

在本論文中，我們提出新的調變式區塊壓縮法來作影像壓縮。我們首先將 影像切割成許多非重疊的區塊，這些區塊被分成四大類，包括：平滑區塊 、可內插還原的區塊、雜亂區塊，和有邊線的區塊。基本的作法是利用不 同的壓縮技術來適應影像區域的特性。我們提出以加權比重的灰度平均值 取代傳統未加權比重的平均值，以便將區塊中的灰度分成兩群。此種方式 所得效果較佳。對於均勻平滑方塊我們可用其灰度平均值來代表；而為了 增加壓縮倍率，部份平滑區塊改以內插法來還原。對於雜亂區塊則採用我 們改進的區塊壓縮法來處理。為了改善影像品質，我們則以三個代表色的 區塊壓縮法來處理含邊線的區塊。在還原影像的視覺品質和方均誤差上， 我們都獲得不錯的實驗結果。當區塊大小是四乘四時，壓縮結果約為每個 影像點需用0.8-1.0 位元。若將向量量化的技巧也用進來，則每個影像點 約需用0.7-0.9 位元。 A new Adaptive Block Truncation Coding (ABTC) is proposed in the thesis for image compression. First, the image is partitioned into non-overlapping blocks that are classified into four cate- gories : the smooth blocks, interpolatable blocks, texture blocks, and edge blocks. The basic idea is to use various coding schemes to allow adaptation to the local image characteristics. A modified BTC using the radius weighted mean (RWM) is proposed, and it is shown to get better performance than most of the re- ported BTC. Besides, we also adopt the interpolation policy to reduce the bit rate. A three- level BTC technique is applied to improve the edge quality in the decoded images. Simulation re- sults show that good visual quality and good MSE performance are obtained in the decoded images. The bit rate is 0.8-1.0 bpp if each block is of size 4x4. If VQ is also used in the proposed ABTC system, then the bit rate becomes 0.7-0.9.