Digital Halftoning and Gamut Mapping for an Inkjet Nail Printer and Digital Halftoning and Descreening with Deep Learning

Baekdu Choi, Purdue University

Abstract

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.Error diffusion algorithms are commonly used for digital halftoning, but they are known to suffer from halftone artifacts. The halftoning algorithm proposed in this dissertation alleviates this issue by adding two features. First, we add a dot-off-dot feature for better halftone texture. Second, we propose a blending-in of the DBS-screened halftone image to the original image as a preprocessing step. These result in halftone images with better texture and reduced artifacts. We demonstrate this by providing a comparison of digitally simulated halftone images.Next, we discuss the gamut mapping algorithm for an inkjet printer, which maps all the colors within the input image to colors reproducible with the printer. Formerly used algorithms for this step suffer from visible desaturation since they only exploit the part of the printer gamut that overlaps with the sRGB gamut. To solve this issue, we add a step we call gamut alignment, which enables the printer to fully exploit the entire printer gamut. We show both the digitally simulated gamut mapped images and the print results from the printer to illustrate the benefits of the added step.We also investigate applying two deep generative models to digital halftoning with the aim of generating halftones with comparable quality to those generated with the DBS algorithm. For the first framework, we apply conditional GANs using two discriminators with different receptive field size and a generator consisting of densely connected blocks. For the second framework, deep autoregressive (AR) models, we propose mapping input images into a feature space using a single forward pass of a deep neural network and then applying a shallow autoregressive model at the end output. Our methods show promising results; halftones generated with our algorithms are less noisy than those generated with a DBS screen and do not contain artifacts commonly associated with error diffusion type algorithms.Lastly, we propose a GAN-based descreening algorithm that generates reconstructed images with realistic texture. Current state-of-the-art descreening algorithms have two issues: first, they mostly are PSNR-oriented reconstruction algorithms, which tend to generate blurry images that lack texture. Furthermore, these algorithms are typically trained with halftone images generated using the Floyd-Steinberg error diffusion algorithm, which is known to generate visible artifacts. To address these issues, first, we propose a GAN-based descreening algorithm that generate images with abundant texture. Next, we propose using the DBS algorithm instead of Floyd-Steinberg error diffusion for generating the halftone images for the training dataset. Both qualitative and quantitative comparisons show that our algorithm outperforms state-of-the-art descreening algorithms significantly.

Degree

Ph.D.

Advisors

Zoltowski, Purdue University.

Subject Area

Artificial intelligence|Mathematics

COinS