Multi-Scale Halftone Image Generation Model Based on Generative Adversarial Network

Digital halftoning algorithms in printers convert continuous-tone images into dot-matrix data by adjusting dot density, and their performance directly affects the visual quality of printed outputs. However, traditional halftoning methods struggle to reconstruct complex textures and preserve high-frequency details. Although deep learning-based halftoning approaches have made notable progress, there remains substantial room for improvement in image quality. To address these challenges, this paper proposes a multi-scale halftone generation model based on generative adversarial networks (GANs), incorporating a novel generator–discriminator architecture. The model effectively converts continuous-tone images into high-quality halftone representations, while preserving complex textures and high-frequency details. Experimental results demonstrate that, compared with traditional and existing deep learning methods, the proposed model consistently achieves performance improvements across texture, smooth, edge and high-frequency detail regions while maintaining computational efficiency.