Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Combined Flicker-banding and Moire Removal for Screen-Captured Images

Libo Zhu, Zihan Zhou, Zhiyi Zhou, Yiyang Qu, Weihang Zhang, Keyu Shi, Yifan Fu, Yulun Zhang

TL;DR

This work tackles the pervasive problem of restoring screen-captured images corrupted by coexisting moiré patterns and flicker-banding. It introduces MIRAGE, a large-scale real-world dataset, and CLEAR, a unified restoration framework that jointly addresses both artifacts. CLEAR combines an ISP-based RAW-domain simulation to diversify degradations, a frequency-domain segmentation front-end to target mid-frequency artifacts, and a trajectory-aligned loss to stabilize diffusion-based restoration, all trained with LoRA for efficiency. Experiments show CLEAR consistently outperforms cascaded baselines and ablations, delivering higher perceptual quality and structural fidelity in realistic capture conditions, with practical implications for mobile photography and screen-capture workflows.

Abstract

Capturing display screens with mobile devices has become increasingly common, yet the resulting images often suffer from severe degradations caused by the coexistence of moiré patterns and flicker-banding, leading to significant visual quality degradation. Due to the strong coupling of these two artifacts in real imaging processes, existing methods designed for single degradations fail to generalize to such compound scenarios. In this paper, we present the first systematic study on joint removal of moiré patterns and flicker-banding in screen-captured images, and propose a unified restoration framework, named CLEAR. To support this task, we construct a large-scale dataset containing both moiré patterns and flicker-banding, and introduce an ISP-based flicker simulation pipeline to stabilize model training and expand the degradation distribution. Furthermore, we design a frequency-domain decomposition and re-composition module together with a trajectory alignment loss to enhance the modeling of compound artifacts. Extensive experiments demonstrate that the proposed method consistently. outperforms existing image restoration approaches across multiple evaluation metrics, validating its effectiveness in complex real-world scenarios.

Combined Flicker-banding and Moire Removal for Screen-Captured Images

TL;DR

This work tackles the pervasive problem of restoring screen-captured images corrupted by coexisting moiré patterns and flicker-banding. It introduces MIRAGE, a large-scale real-world dataset, and CLEAR, a unified restoration framework that jointly addresses both artifacts. CLEAR combines an ISP-based RAW-domain simulation to diversify degradations, a frequency-domain segmentation front-end to target mid-frequency artifacts, and a trajectory-aligned loss to stabilize diffusion-based restoration, all trained with LoRA for efficiency. Experiments show CLEAR consistently outperforms cascaded baselines and ablations, delivering higher perceptual quality and structural fidelity in realistic capture conditions, with practical implications for mobile photography and screen-capture workflows.

Abstract

Capturing display screens with mobile devices has become increasingly common, yet the resulting images often suffer from severe degradations caused by the coexistence of moiré patterns and flicker-banding, leading to significant visual quality degradation. Due to the strong coupling of these two artifacts in real imaging processes, existing methods designed for single degradations fail to generalize to such compound scenarios. In this paper, we present the first systematic study on joint removal of moiré patterns and flicker-banding in screen-captured images, and propose a unified restoration framework, named CLEAR. To support this task, we construct a large-scale dataset containing both moiré patterns and flicker-banding, and introduce an ISP-based flicker simulation pipeline to stabilize model training and expand the degradation distribution. Furthermore, we design a frequency-domain decomposition and re-composition module together with a trajectory alignment loss to enhance the modeling of compound artifacts. Extensive experiments demonstrate that the proposed method consistently. outperforms existing image restoration approaches across multiple evaluation metrics, validating its effectiveness in complex real-world scenarios.
Paper Structure (28 sections, 42 equations, 54 figures, 5 tables)

This paper contains 28 sections, 42 equations, 54 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset: MIRAGE
  4. Data Collection
  5. Data Capture
  6. Methods
  7. ISP-based Simulation Pipeline (ISPS)
  8. Frequency Domain Segmentation (FS)
  9. Trajectory-aligned Loss (TA)
  10. Overall Training Objective.
  11. Experiments
  12. Experiments Setup
  13. Comparison with combined models
  14. Main Results
  15. Ablation Study
  16. ...and 13 more sections

Figures (54)

  • Figure 1: Overview of our dataset (MIRAGE) and model (CLEAR) results. Left: examples of three types of degradations in our dataset: flicker-banding, moiré, and flicker-banding&moiré. Right: the flicker-banding&moiré removal results of our proposed model on real-world images compared with Low-Quality (LQ) and Ground-Truth (GT) images.
  • Figure 2: Overview of the simulation pipeline. Given an sRGB image with moiré, Inverse-ISP model is used to convert it to RAW format. Then, a flicker-banding mask is generated and applied in the RAW domain to simulate banding artifacts. Finally, the ISP model converts the degraded RAW image back to RGB space, producing a training pair with both moiré and banding artifacts.
  • Figure 3: Example of simulation banding types. 1: simple; 2: diamond; 3: curve; 4: cracked; 5: complex. More examples are in the supplementary material.
  • Figure 4: Overview of the proposed CLEAR framework. (a) Frequency-domain segmentation (FS) module. (b) The overall CLEAR architecture. (c) The training objective combines trajectory alignment loss (TA) with perceptual and pixel losses.
  • Figure 5: Visual comparison with combined methods.
  • ...and 49 more figures