Color image quantization & halftoning based on human visual perception

Abstract

In this dissertation, we propose two algorithms that address two important issues in color quantization. The first one being the incorporation of spatial or contextual information of an input image to the process of color quantization, thus allowing quantization to give priorities to different regions of the image and focus on the regions of the image having important color information. In considering the way the human visual system deals with color difference, the second algorithm is concerned with color space. The algorithm is a simple but effective color quantization technique, which performs 3D frequency diffusion in RGB color space. By manipulating a diffusion filter, a cross-space operation is achieved, obtaining quantization effect in a color space without transforming the image to that space. For most color displays, processing of colors is performed on the R, G and B components. The requirement for human vision system, however, would favour a different scheme because the Euclidean distance in the RGB color space is not consistent with the way the human visual system deals with color difference. Theoretically, a luminance-chrominance color space that correlates with human color perception is more suitable to be adopted than the RGB for many applications. The transformation processes between color spaces is, however, bound up with computational errors. As a result, the overall performance of any color quantization scheme that involves conversion between color spaces may not be better than a quantization completely in RGB. This dissertation begins with a brief history of color quantization and halftone development as well as descriptions of major techniques. Followed by that, the essential color knowledge encountered in the context of color image quantization, as well as the fundamental concepts, models, metrics and their formulation for evaluating the quality of quantization outputs is examined. Experimental results show that our algorithms perform better in some aspects than two popularly used algorithms for color quantization. Finally, description of the algorithms representing the core of the research along with in depth discussions of each algorithm will be given.