應用於顯示影像的再量化法研究

Abstract

影像必須經過再量化的處理才能顯示在只能處理少許灰度值或色彩的低階 繪圖設備上。本篇論文包含彩色與灰度影像的再量化處理。第一、我們介 紹新的量化技巧稱為基數分配。當基數分配作為灰度影像再量化法時，相 對於常用的序顫法、誤差分散法、與斑點分散法有許多優點。在本文中， 我們提出一種相似量度叫作量化誤差曲線，來強調基數分配法其中一項優 點。第二、我們更進一步地拓展基數分配在彩色影像再量化的應用，由此 來克服個人電腦顯示彩色影像的困難。第三、有時候為了能使細微的部份 能顯現清楚，量化影像需進一步的強化。在本篇論文中，我們提出一些新 的演算法來強化量化過的影像。首先我們引薦反模糊的觀念來改善誤差分 散法。接著，我們為所有量化演算法提出改進，使它們能夠捕捉灰度量化 影像的細微部份。最後，我們也為彩色量化影像提出一套新的強化方法。 第四，也就是本文的最後一部份，我們應用基數分配技巧來壓縮影像。基 數分配法吸引人的優點中有一項是能夠產生序顫紋理，此紋理非常適用在 影像壓縮上。在本文中，我們藉由基數分配來壓縮影像，並與多階序顫法 比較。 An image must be re-quantized when it will be displayed on a low level graphics device, i.e., a graphics device that can handle only few gray levels or colors. This thesis addresses the problems of both grayscale and color image requantizations. First, we introduce a new technique called cardinality distribution. As a grayscale image requantization method, cardinality distribution has many advantages over the popular algorithms, ordered dither, error diffusion, and dot diffusion algorithms. To classify one of the advantages for cardinality distribution, we will propose a similarity measure, called the quantization error curve. It is useful to measure the similarity of requantized images. Second, we extend the application of cardinality distribution to color image requantization. When cardinality distribution is applied to display color images, it can overcome the difficulties of displaying color images on personal computers. Third, for visual effect, requantized images are sometimes enhanced. In this thesis, we propose some enhancement techniques for requantized images. At first, we employ the concept of defuzzification to improve the error diffusion algorithm. Next we propose a modification for all quantization algorithms to capture the sharp details in the requantized grayscale images. Finally, we propose a new enhancement algorithm for requantized color images. At last, we employ cardinality distribution to compress images. A very interesting property of cardinality distribution is that it can produce dithering patterns which are suitable for image compression. In this thesis, we compress images by cardinality distribution and compare the results with those by multilevel dither technque.