A real-time scalable color quantizer trainer/encoder

Abstract

In many color-imaging applications, it is desirable to display an image with as few different colors as possible with minimal loss in image quality. While good image quality is achievable using traditional Vector Quantization techniques. they are too slow for real-time vidm applications. An architectural design of a real-time, scalable color quantizer architecture is presented. It implements our fast Tree Structure Vector Quantization algorithm with a variable-size cubical prequantizer based on human perception proposed earlier. The design is scalable and uses different configurations of the processing and memory elements to process any 24-bit to 72-bit colors per input pixel to produce a 8-bit to 24-bit color palette.