影像的調色盤設計，壓縮及隱藏

Abstract

本論文中，我們先利用半徑加權中心(RWM)的特性，提出RWM-cut這個快速的資料二分工具，以便能快速且低失真地執行色彩量化。除了三度空間的作法，我們也提出一度空間的版本，以便執行即時的色彩量化。另外，我們也提出三個改良型的BTC式影像壓縮法，分別是RBTC，EBTC以及HABTC (或HABTC-VQ) 壓縮法。其中RBTC與EBTC是用來提昇BTC在固定壓縮倍率下之PSNR值，而HABTC則是在浮動壓縮倍率下，用來改善BTC的效能。最後，我們以區塊基底導向(Base-oriented)的法則，提出一個簡單且有效率的隱像技巧。此新技巧不須對隱密影像(security image)或主体影像(host image)做前置處理，以及在接收側欲從合成影像(mixed image)中取出隱密影像時，也不須使用到對照表或原本的主体影像。

In this dissertation we first proposed a fast color quantization method called the RWM-cut method which uses the good splitting ability of radius weighted mean (RWM). Besides the 3-dimensional version, a 1-dimensional version of RWM-cut method is also presented for real-time applications. Experiments show that the proposed method is feasible and visually acceptable. We then relate RWM to BTC compression method to obtain several new BTC schemes (including RBTC, EBTC, and HABTC (or HABTC-VQ) coding schemes) so that the compression efficiency of BTC can be improved. Both RBTC and EBTC are useful in improving the PSNR performance of BTC at fixed bit rate, while the HABTC (or HABTC-VQ) improves BTC at variable bit rates. Finally, a simple and effective method called the base-oriented algorithm is proposed for image hiding. The proposed hiding method does not need to preprocess either the security image or host image. When the security image is to be extracted from the mixed image, the extraction is easy, and neither the original host image nor the look-up table is needed.