Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Guiding WaveMamba with Frequency Maps for Image Debanding

Xinyi Wang, Smaranda Tasmoc, Nantheera Anantrasirichai, Angeliki Katsenou

TL;DR

This work tackles banding artifacts caused by low-bitrate compression by introducing a post-processing debanding method that blends WaveMamba with a frequency masking map to preserve textures. It details a three-variant masking strategy integrated into a wavelet-state-space architecture and benchmarks it against open-source debanding methods on deepDeband and BAND-2k datasets. The results show strong PSNR and low DBI for the WaveMamba variants, with WaveMamba-MAP offering the best perceptual preservation in many cases, while revealing misalignments between traditional banding metrics and human perception. The study highlights the importance of perceptually oriented evaluation for debanding and suggests future work on developing such metrics.

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.

Guiding WaveMamba with Frequency Maps for Image Debanding

TL;DR

This work tackles banding artifacts caused by low-bitrate compression by introducing a post-processing debanding method that blends WaveMamba with a frequency masking map to preserve textures. It details a three-variant masking strategy integrated into a wavelet-state-space architecture and benchmarks it against open-source debanding methods on deepDeband and BAND-2k datasets. The results show strong PSNR and low DBI for the WaveMamba variants, with WaveMamba-MAP offering the best perceptual preservation in many cases, while revealing misalignments between traditional banding metrics and human perception. The study highlights the importance of perceptually oriented evaluation for debanding and suggests future work on developing such metrics.

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.

Paper Structure

This paper contains 12 sections, 10 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. WaveMamba Network
  4. Frequency Masking Map
  5. Experimental Design
  6. Datasets
  7. Evaluation Metrics
  8. Benchmarking Methods
  9. Results and Discussion
  10. Performance Comparison
  11. Visual Inspection
  12. Conclusion and Future Work

Figures (2)

  • Figure 1: Visual comparison of debanding performance during inference on the BAND-2k dataset. Each row represents a difficulty level: (a) easy, (b) medium, and (c) difficult based on DBI values. Columns correspond to: Banded Input, WaveMamba variants, and Pristine.
  • Figure 2: Qualitative comparisons of the models trained and tested on deepDeband patches.