It Takes Two: A Duet of Periodicity and Directionality for Burst Flicker Removal
Authors: Lishen Qu, Shihao Zhou, Jie Liang, Hui Zeng, Lei Zhang, Jufeng Yang
Organizations: Nankai International Advanced Research Institute (SHENZHEN·FUTIAN) · College of Computer Science, Nankai University · OPPO Research Institute · The Hong Kong Polytechnic University · Peng Cheng Laboratory
Flicker artifacts, arising from unstable illumination and row-wise exposure inconsistencies, pose a significant challenge in short-exposure photography, severely degrading image quality. Unlike typical artifacts, e.g., noise and low-light, flicker is a structured degradation with specific spatial-temporal patterns, which are not accounted for in current generic restoration frameworks, leading to suboptimal flicker suppression and ghosting artifacts. In this work, we reveal that flicker artifacts exhibit two intrinsic characteristics, periodicity and directionality, and propose Flickerformer, a transformer-based architecture that effectively removes flicker without introducing ghosting. Specifically, Flickerformer comprises three key components: a phase-based fusion module (PFM), an autocorrelation feed-forward network (AFFN), and a wavelet-based directional attention module (WDAM). Based on the periodicity, PFM performs inter-frame phase correlation to adaptively aggregate burst features, while AFFN exploits intra-frame structural regularities through autocorrelation, jointly enhancing the network's ability to perceive spatially recurring patterns. Moreover, motivated by the directionality of flicker artifacts, WDAM leverages high-frequency variations in the wavelet domain to guide the restoration of low-frequency dark regions, yielding precise localization of flicker artifacts. Extensive experiments demonstrate that Flickerformer outperforms state-of-the-art approaches in both quantitative metrics and visual quality. The source code is available at https://github.com/qulishen/Flickerformer.
Figures & tables
Figure 1: Motivation of Flickerformer. (a) Swapping the phase components between two consecutive flickering images leads to an exchange of flicker patterns, indicating that phase encodes the spatial distribution of flicker. (b) Illustration of the intrinsic flicker characteristics and the corresponding module designs: PFM and AFFN are devised based on periodicity, while WDAM is inspired by directionality.
Figure 2: Overview of Flickerformer. (a) The proposed Flickerformer adopts an asymmetric U-shaped encoder–decoder architecture. Following the previous work [ 55 ] , the input consists of three frames, while the output is a single restored frame. (b) The proposed phase fusion module (PFM), which performs feature fusion based on phase correlation. The operation of the phase spectral filtering (PSF) corresponds to Eqs. 2 and 3 , and the fusion module corresponds to Eq. 5 . (c) The proposed wavelet-based directional attention module (WDAM), including a low-frequency attention branch and a high-frequency branch to capture directional features.
Figure 3: The mechanism of AFFN. The “conj” means conjugate operation. Unlike PFM, the proposed AFFN performs correlation within the same feature representation, which is referred to as autocorrelation.
Figure 4: Feature visualization. (a) The non-reference frame provides both clean and flicker-corrupted regions to guide the base frame. (b) Without the proposed PFM, the flicker and normal regions are treated almost equally during feature fusion. (c) Benefiting from phase correlation, PFM performs an effective pre-filtering step to distinguish features in the reference frame.
Figure 5: Visualization of wavelet decomposition. The high-frequency components in LH (or HL) bands often capture regions with sharp luminance variations, which helps distinguish flickering areas from naturally dark regions and guides the attention mechanism to emphasize flicker-affected regions.
Figure 6: Qualitative comparison of flicker removal methods. Our Flickerformer achieves superior visual quality, effectively removing flicker while preserving fine textures and color. The blue background indicates the difference map between the result and the ground truth, where pixels exceeding a certain threshold are highlighted in red.
Method
PSNR ↑
SSIM ↑
LPIPS ↓
Params (M)
Flops (G)
Stripformer [ 60 ]
29.223
0.892
0.058
19.71
681.64
Uformer [ 63 ]
30.544
0.910
0.056
18.12
145.24
Restormer [ 72 ]
30.630
0.917
0.055
26.10
141.16
HDRTransformer [ 45 ]
30.031
0.918
0.054
1.04
272.12
Retinxformer [ 5 ]
29.598
0.899
0.055
3.74
184.14
FFTformer [ 32 ]
29.478
0.895
0.050
14.88
131.71
Table 1: Quantitative comparison of 16 methods on the benchmark dataset [ 55 ] . The best and second-best scores are highlighted in bold and underlined , respectively.
Models
FFN [ 41 ]
LeFF [ 63 ]
GDFN [ 72 ]
FRFN [ 76 ]
AFFN Ours
Params (M)
4.45
4.60
3.92
4.03
3.92
Flops (G)
139.31
146.73
128.76
128.76
128.76
PSNR (dB)
31.876
30.954
30.959
30.961
31.226
Table 2: Ablation study of using different feature refinement feed- forward network.
Models
Swin SA [ 41 ]
Top-k SA [ 9 ]
Condensed SA [ 75 ]
ASSA [ 76 ]
WDAM Ours
Params (M)
3.92
3.96
3.46
3.92
3.92
Flops (G)
139.36
145.20
132.29
139.42
128.76
PSNR (dB)
30.896
30.894
30.981
30.997
31.226
Table 3: Ablation study of using different attention mechanisms.
Figure 7: Visualization comparison of different feature refinement feed-forward networks. Although both FRFN and AFFN can restore flicker-affected regions ( red box), FRFN tends to introduce ghosting artifacts ( yellow box), as it struggles to distinguish between motion variations and flicker variations.
Figure 8: Visualization comparison of different attention modules. Benefiting from directional attention and high-frequency guidance, WDAM can more accurately identify flicker-affected regions and achieve more thorough flicker removal.
CNN
PFM
FRFN
AFFN
ASSA
MDAM
PSNR ↑
SSIM ↑
(a)
✔
✔
✔
30.449
0.912
(b)
✔
✔
✔
30.728
0.914
(c)
✔
✔
✔
30.831
0.915
(d)
✔
✔
✔
30.822
0.915
(e)
✔
✔
✔
31.226
0.920
Table 4: Quantitative evaluations of each module of Flickerformer. The CNN, FRFN, and ASSA modules are components in AST [ 76 ] , while the PFM, AFFN, and MDAM are the corresponding modules in the proposed method.
Figure 9: Example of the limitation. Flickerformer struggles to restore regions affected by large-scale light extinction.
Flicker-banding (FB), arises from temporal aliasing between a camera's rolling shutter and a display's brightness modulation, degrading screen-captured image readability with color shifts and jagged patterns. Existing single-frame methods with simplified parametric stripe models cannot reliably distinguish these artifacts from genuine texture. To address this, we conduct a systematic analysis of complex FB morphologies and reveal their significant variation across exposure settings, motivating a multi-frame bracketed RAW restoration paradigm. We construct Bricker, a synthetic-real bracketed RAW dataset built via ray-tracing-based physical simulation and automated multi-exposure capture tool. We further propose BRACE: Bracketed RAW Flicker-Banding Removal, a multi-frame restoration model that utilizes frequency-aware banding prior and a multi-scale spatial cross-attention modulator (MSCAM) for cross-exposure spatial fusion. We also introduce the Stripe Frequency Consistency (SFC) metric to evaluate banding removal. Experiments demonstrate state-of-the-art performance on both synthetic and real benchmarks. Our dataset and code are available at: https://github.com/ZZH-qwq/BRACE.
Capturing screens is now routine in our everyday lives. But the photographs of emissive displays are often influenced by the flicker-banding (FB), which is alternating bright%u2013dark stripes that arise from temporal aliasing between a camera's rolling-shutter readout and the display's brightness modulation. Unlike moire degradation, which has been extensively studied, the FB remains underexplored despite its frequent and severe impact on readability and perceived quality. We formulate FB removal as a dedicated restoration task and introduce Removal of Image Flicker-Banding via Latent Diffusion Enhancement, RIFLE, a diffusion-based framework designed to remove FB while preserving fine details. We propose the flicker-banding prior estimator (FPE) that predicts key banding attributes and injects it into the restoration network. Additionally, Masked Loss (ML) is proposed to concentrate supervision on banded regions without sacrificing global fidelity. To overcome data scarcity, we provide a simulation pipeline that synthesizes FB in the luminance domain with stochastic jitter in banding angle, banding spacing, and banding width. Feathered boundaries and sensor noise are also applied for a more realistic simulation. For evaluation, we collect a paired real-world FB dataset with pixel-aligned banding-free references captured via long exposure. Across quantitative metrics and visual comparisons on our real-world dataset, RIFLE consistently outperforms recent image reconstruction baselines from mild to severe flicker-banding. To the best of our knowledge, it is the first work to research the simulation and removal of FB. Our work establishes a great foundation for subsequent research in both the dataset construction and the removal model design. Our dataset and code will be released soon.
Libo Zhu, Zihan Zhou, Xiaoyang Liu +4
Shanghai Jiao Tong University · Central Media Technology Institute, Huawei
Photography is the art of painting with light, yet nighttime scenes are shaped by competing degradations: intense flares obscure scene structure, while photon-limited regions collapse into noise. Conventional approaches address these factors in isolation, overlooking the fact that these degradations are fundamentally entangled. To bridge this gap, we introduce LUCID, a unified framework that reframes nighttime restoration as a continuous and controllable process rather than a fixed correction. We decompose nighttime restoration into two cooperative components: a flare disentanglement module that lifts the 'curtain' of optical artifacts to provide reliable structural guidance, and a diffusion-driven module that leverages generative priors to reconstruct clean and well-exposed imagery. Crucially, LUCID introduces explicit controllability through a novel four-mode training strategy, enabling users to steer the restoration process via classifier-free guidance (CFG) and allowing selective control over light sources and their associated flare and ghosting artifacts, while also supporting high dynamic range (HDR) reconstruction through continuous exposure control. Extensive experiments demonstrate that LUCID consistently outperforms state-of-the-art methods across diverse real-world nighttime scenarios.