Image/Video Demoiréing

Adaptive Video Demoiréing Network with Subtraction-Guided Alignment

Fig. 1. AVDNet architecture.

Moiré artifacts arise from frequency interference between display grids and camera sensors, causing visually disturbing patterns in captured photos and videos. We address both image and video demoiréing by exploiting spectral and temporal characteristics of moiré patterns. Our adaptive video demoiréing network (AVDNet) suppresses moiré adaptively in the implicit frequency domain via a learnable bandpass filter (ABB), and uses inter-frame subtraction maps to guide temporal alignment and prevent moiré propagation across frames (SGAB). For image demoiréing, we propose a multiscale coarse-to-fine strategy that exploits correlations between moiré frequencies at multiple scales.

Visual comparison of demoiréing results — Fig. 2. Qualitative comparison on VDmoire dataset: (a) Input, (b) VDmoire, (c) DTNet, (d) Ours, (e) GT.

TABLE I. Quantitative comparison on VDmoire dataset. **Bold**: best, underline: 2nd best.
Method	TCL-V1			TCL-V2			iPhone-V1			iPhone-V2			Complexity
Method	PSNR↑	SSIM↑	LPIPS↓	PSNR↑	SSIM↑	LPIPS↓	PSNR↑	SSIM↑	LPIPS↓	PSNR↑	SSIM↑	LPIPS↓	Params(M)	TFLOPs	Runtime(ms)
DMCNN	20.321	0.703	0.321	20.707	0.793	0.385	21.967	0.712	0.280	21.416	0.749	0.496	1.43	0.736	32.0
WDNet	19.650	0.726	0.289	20.334	0.847	0.288	19.818	0.722	0.300	20.613	0.832	0.297	3.92	3.100	63.1
ESDNet	22.026	0.734	0.199	24.898	0.874	0.163	22.537	0.751	0.218	25.064	0.853	0.165	5.93	4.484	43.6
VDmoire	21.725	0.733	0.202	23.460	0.857	0.163	21.990	0.707	0.221	25.230	0.860	0.157	5.98	1.216	650.2
DTNet	24.119	0.801	0.163	26.153	0.877	0.128	24.821	0.794	0.172	26.593	0.854	0.149	7.36	1.174	620.3
DTCINet	21.881	0.744	0.181	-	-	-	25.381	0.876	0.148	-	-	-	5.17	-	-
STD-Net	22.075	0.740	0.196	20.705	0.834	0.197	21.889	0.717	0.208	24.216	0.842	0.180	21.77	1.056	151.7
FPANet	21.953	0.784	0.173	-	-	-	-	-	-	25.446	0.883	0.146	68.89	-	-
Ours	24.730	0.819	0.140	26.478	0.877	0.128	25.492	0.810	0.148	26.593	0.855	0.131	4.16	0.540	52.3

Publications

Seung-Hun Ok, Young-Min Choi, Seung-Wook Kim, and Se-Ho Lee, “Adaptive video demoiréing network with subtraction-guided alignment,” IEEE Signal Processing Letters, vol. 32, Jul. 2025. [DOI] [Code]
Duong Hai Nguyen, Se-Ho Lee, and Chul Lee, “Multiscale coarse-to-fine guided screenshot demoiréing,” IEEE Signal Processing Letters, vol. 30, pp. 898–902, Jul. 2023. [DOI]