數位彩色影像之量化與重建

Abstract

彩色影像量化是期望在人眼沒有感受到差異的情況下，將彩色影像由較多的色彩數量簡化為較少的表示色彩數量。彩色影像量化對於輸出能顯示的色彩數量不足時，是一個很有用的方法。此外它也可以節省儲存空間和傳輸時間。因此，彩色影像量化在現在是一個很重要的議題。 在這篇論文中，我們將會介紹四種影像量化的演算法。前兩種是廣為人知的方法：中位數切割法和中心點切割法。這兩種方法的切割平面是由平衡數值總類的數量來決定。第三種方法藉由計算在像素數值鄰近點的距離的量化法，主要希望可以使量化誤差達到最小。第四種方法是藉由保持量化前後色彩分佈不變的量化法。藉由保持量化前後色彩分佈不變的特性，我們還可以做其他的應用。 最後會介紹一種不需要知道影像量化方法的重建方法。期望在視覺上可以將量化過後的影像重建回沒量化過，主要是可以消除量化時產生的影像輪廓。

The color image quantization expects that a color image can use less number of representative colors to represent it and the human eyes can not notice. It is helpful when the output do not have enough colors to display. Also, it can save storage memory and the transmit time. Thus the color image quantization is an important issue in nowadays. In this thesis, we will introduce four kinds of image quantization algorithms. The pervious two are well-know: the median-cut and the center-cut algorithm. The cutting plane is decided by balancing populations. The third one is called quantization using distances between adjacent colors algorithm which wishes to minimal the total quantization errors. The Fourth one is quantization by preserving color distribution features, which want to keep the color distribution unchanged. By preserving the same color distributions, we also do some applications. Last, we will illustrate a color image dequantization algorithm which is blind. It will improve the color contours and make the quantized color image looks like an original one.