Multi-Scale Frequency VAE-GAN: A Universal Approach for Inverting Diverse Halftone Images
Year: 2025
Type: article
Abstract: Inverse halftoning aims to reconstruct continuous-tone images from halftone patterns. Although many learning-based methods have achieved promising results, they often struggle to generalize across diverse halftoning schemes due to variations in noise distribution within the frequency domain. To alleviate this issue, this paper presents a multi-scale frequency-aware VAE-GAN framework that enhances the reconstruction of various halftone types. The model consists of a content synthesis module based on a variational autoencoder and a detail refinement stage using a generative adversarial network. Frequency-aware feature extraction and adaptive fusion mechanisms are employed to better capture multi-scale structures and suppress halftone artifacts. Experimental evaluations on several halftone datasets indicate that the proposed method provides consistent and visually faithful restoration, achieving competitive quantitative and perceptual metrics compared with existing approaches. While the framework builds upon prior multi-stage designs, it offers a practical and extensible solution for general inverse halftoning tasks rather than a radical architectural innovation.
Source: IEEE Access
Authors: Zhanfeng Li, Hongwu Zhan, Yinwei Zhang, Shaojie Tang, Yunyun Shen
Institution: Zhejiang University of Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,075
OpenAlex ID: W4416714569