A New Method of Halftoning and Inverse Halftoning Based on GAN Network

Halftoning is a method of quantifying a continuous tone image into a binary image. Halftone methods can be categorized into dithering, error diffusion and iterative method. We present a new method different from traditional halftoning algorithms for learning the mapping between continuous images and halftone images using conditional generative adversarial networks (conditional GANs). We regard halftoning and inverse halftoning as the process of image-to-image translation, and use the classic pix2pixHD network. In this work, we use multi-scale generator and discriminator architectures to perform both halftoning and its structural reconstruction. The experimental results show that this method can better fit some classical halftoning algorithms to realize halftoning and inverse halftoning. Compared with the existing methods, our method can better realize reconstruction of halftone images.