Quantization of color images for display/printing on limited color output devices

Abstract

A large number of output devices in use today are either bilevel or can produce only a limited number of display levels (gray-scale or color). Most color graphics terminals conforming to Enhanced Graphics Adapter (EGA), Professional Graphics Adapter (PGA), or Video Graphics Array (VGA) standards can display from 16–256 colors, whereas real-world (externally acquired) images constitute typically 16M colors. In this paper, a new color quantization algorithm has been proposed which maps an original image into an output image with a limited number of colors, while still preserving the image quality. The algorithm itself is based on the concepts of vector quantization where a color vector is defined by red, green, and blue components and, based on a random sampling of the input image, a color mapping table is generated. The random sampling provides an estimate of the color distribution of the input image, which is then further combined by a clustering technique to derive the desired number of output colors. A mapping process results in a limited-color output image which is optionally preprocessed (in cases where the number of output colors is very small) by a pseudo-random dithering algorithm rendering a high-quality output. This postprocessing step is particularly useful in images with very few output colors, e.g., 16. Through examples, it is shown that input images with over 16M colors can be easily displayed in as few as 16 colors, with negligible degradation in quality.