Hybrid approaches to color image quantization

Color image quantization is the irreversible transformation of a truecolor image into a color-mapped image consisting of K carefully selected representative colors. There are many possible mappings of a truecolor image to a quantized image using K colors. The primary goal of color quantization is to minimize the visual distortion between the original image and the quantized image.; This dissertation proposes a heterogeneous-cut algorithm that combines the speed of oct-cut methods with the accuracy of 24-bit precision, variance-minimization and principal component oblique-cut methods to achieve high quality quantized images quickly. This dissertation also presents a fast and effective (improves image quality) method for generalizing activity weighting to any histogram-based color quantization algorithm. The value of the heterogeneous-cut algorithm and activity weighting is validated by a comprehensive empirical analysis of thirty-nine other hierarchical color quantization techniques using a test set consisting of twenty-five diverse images. Of the high quality quantization techniques studied in the analysis, the proposed heterogeneous-cut algorithm is the fastest.; This dissertation shows that the maximum intercluster distance is not an appropriate error measure for color image quantization (MinMax). Furthermore, this dissertation proposes a new non-hierarchical color quantization technique called weighted MinMax that is a hybrid between the MinMax and Linde-Buzo-Gray (LBG) algorithms. The new method incorporates frequency (or activity weighting) information in order to obtain high quality quantized images with significantly less visual distortion than the MinMax algorithm. However, the running time of both the MinMax and the weighted MinMax algorithm is not competitive with any of the hierarchically divisive methods.