<p>In this dissertation, we propose four novel digital image processing algorithms. First, we discuss a novel digital halftoning algorithm that efficiently removes halftone artifacts commonly associated with error diffusion while adding only an insignificant computational cost. Second, we propose a novel gamut mapping algorithm that utilizes the entire printer gamut resulting in more saturated print results. Third, we propose two digital halftoning algorithms using deep neural networks that generate halftones with quality comparable to those generated with the direct binary search (DBS) algorithm. Lastly, we propose a descreening algorithm based on generative adversarial networks (GAN) framework that generates images with realistic texture.</p>

FM halftoning is increasingly popular with traditional analog offset lithographic printing processes. There is a desire to offer this capability with digital presses based on electrophotographic printing (EP) technologies. However, the inherent instability of the EP process challenges the achievement of satisfactory print quality with dispersed-dot, aperiodic halftoning. The direct binary search (DBS) algorithm is widely considered to represent the gold standard of dispersed-dot, aperiodic halftone image quality. In this paper, we continue our previous efforts to adapt DBS to use with the Indigo liquid EP printing technology. We describe a complete color management pipeline for halftoning with a PARAWACS matrix designed using DBS. For the first time, we show actual printed patches obtained using our process. Our gamut mapping is performed in the YyCxCz color space, and is image-dependent. It incorporates several stages of alignment between the input and output spaces, as well as several stages of compression. After the gamut mapping, we tessellate the output color space into six global tetrahedra that each share the neutral axis, as an edge. Then, we determine the Neugebauer Primary Area Coverage (NPAC) for each pixel in the image to be printed by tetrahedral interpolation from the four nearest neighbors in the inverse printer mapping table. These four nearest neighbors are chosen so that only four Neugebauer primaries are used to render each pixel.