Guiding WaveMamba with Frequency Maps for Image Debanding

2025-08-15

AI Summary

Low-bitrate image compression frequently introduces pronounced banding artifacts in smooth regions (e.g., skies), with visual distortion further exacerbated under repeated transcoding—common in user-generated content. To address this, we propose a wavelet state space model guided by a frequency-aware mask map: images are decomposed via wavelet transform, and a learnable frequency mask dynamically prioritizes low-frequency subbands most susceptible to banding; the WaveMamba architecture then enables precise frequency-domain modeling and reconstruction. Our method effectively suppresses banding in smooth areas while preserving high-frequency textures. On the BAND-2k benchmark, it achieves a Banding Index (DBI) of 0.082—significantly outperforming state-of-the-art methods. Comprehensive visual evaluation confirms superior perceptual realism and improved fidelity–detail trade-off compared to existing approaches.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Learning on the Edge & Model Compression
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Compression at low bitrates in modern codecs often introduces banding artifacts, especially in smooth regions such as skies. These artifacts degrade visual quality and are common in user-generated content due to repeated transcoding. We propose a banding restoration method that employs the Wavelet State Space Model and a frequency masking map to preserve high-frequency details. Furthermore, we provide a benchmark of open-source banding restoration methods and evaluate their performance on two public banding image datasets. Experimentation on the available datasets suggests that the proposed post-processing approach effectively suppresses banding compared to the state-of-the-art method (a DBI value of 0.082 on BAND-2k) while preserving image textures. Visual inspections of the results confirm this. Code and supplementary material are available at: https://github.com/xinyiW915/Debanding-PCS2025.

Problem

• Compression at low bitrates causes banding artifacts
• Banding artifacts degrade visual quality in smooth regions
• Proposes a method to restore banding using Wavelet State Space Model

Innovation

• Wavelet State Space Model for banding restoration
• Frequency masking map preserves high-frequency details
• Benchmarking open-source banding restoration methods