Learning Moiré Pattern Elimination in Both Frequency and Spatial Domains for Image Demoiréing

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Mobile sensing devices frequently capture images and videos, but moiré patterns degrade visual quality.
• Moiré patterns arise from interference between digital screens or regular patterns, posing a significant challenge in image acquisition.

Purpose of the Study:

• To propose a novel end-to-end image demoiré method for eliminating complex moiré patterns.
• To enhance image quality degraded by moiré artifacts in mobile sensing applications.

Main Methods:

• A hybrid approach combining frequency and spatial domain processing for moiré pattern elimination.
• Utilizing wavelet transform in the frequency domain to decompose multi-scale image features for precise moiré feature identification.
• Introducing a spatial domain demoiré block (SDDB) to extract and subtract moiré features from mixed signals.

Main Results:

• The proposed method demonstrates superior performance compared to existing state-of-the-art techniques.
• Ablation studies using Grad-CAM confirm the effectiveness of both frequency and spatial domain processing blocks.
• The combined approach significantly enhances moiré feature recognition and elimination capabilities.

Conclusions:

• The novel end-to-end image demoiré method effectively addresses complex moiré patterns.
• The integration of frequency and spatial domain processing offers a robust solution for improving image quality in mobile sensing.
• The proposed technique shows significant potential for practical applications where moiré artifacts are prevalent.