Freqformer: Image-Demoir'eing Transformer via Efficient Frequency Decomposition

2025-05-25

AI Summary

Moiré pattern removal suffers from strong coupling between texture distortion and color distortion, posing a fundamental modeling challenge. This paper proposes the first frequency-domain decoupling Transformer framework, explicitly decomposing moiré artifacts into high-frequency local texture distortions and low-frequency global chromatic shifts, and employing a dual-branch collaborative network to model them separately. We introduce a novel learnable Frequency Composition Transform (FCT) module for adaptive fusion of frequency-domain features, and integrate a Spatial-Aware Channel Attention (SA-CA) mechanism to precisely localize and refine moiré-sensitive regions. By synergistically combining multi-scale wavelet frequency decomposition with a lightweight Transformer architecture, our method achieves state-of-the-art performance across multiple benchmarks. It exhibits superior efficiency—requiring fewer parameters and enabling fast inference—while maintaining high restoration fidelity. The source code is publicly available.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Image demoir'eing remains a challenging task due to the complex interplay between texture corruption and color distortions caused by moir'e patterns. Existing methods, especially those relying on direct image-to-image restoration, often fail to disentangle these intertwined artifacts effectively. While wavelet-based frequency-aware approaches offer a promising direction, their potential remains underexplored. In this paper, we present Freqformer, a Transformer-based framework specifically designed for image demoir'eing through targeted frequency separation. Our method performs an effective frequency decomposition that explicitly splits moir'e patterns into high-frequency spatially-localized textures and low-frequency scale-robust color distortions, which are then handled by a dual-branch architecture tailored to their distinct characteristics. We further propose a learnable Frequency Composition Transform (FCT) module to adaptively fuse the frequency-specific outputs, enabling consistent and high-fidelity reconstruction. To better aggregate the spatial dependencies and the inter-channel complementary information, we introduce a Spatial-Aware Channel Attention (SA-CA) module that refines moir'e-sensitive regions without incurring high computational cost. Extensive experiments on various demoir'eing benchmarks demonstrate that Freqformer achieves state-of-the-art performance with a compact model size. The code is publicly available at https://github.com/xyLiu339/Freqformer.

Problem

• Addressing complex moiré pattern artifacts in images
• Improving frequency-aware demoiréing with targeted separation
• Enhancing reconstruction fidelity via adaptive frequency fusion

Innovation

• Transformer-based frequency separation for demoiréing
• Dual-branch architecture handles texture and color
• Learnable Frequency Composition Transform for fusion