Texture and Noise Dual Adaptation for Infrared Image Super-Resolution

2023-11-15

AI Summary

In infrared image super-resolution, visible-light guidance often introduces noise and blurring artifacts. To address this, we propose Domain-Adaptive SRGAN (DASRGAN), the first framework featuring dual-path, target-oriented domain adaptation: a texture-guided path leveraging prior-extracted discriminators and Sobel-edge–guided adversarial loss for enhanced texture alignment, and a noise-guided path incorporating an explicit noise-adversarial loss to decouple and suppress noise transfer. By jointly integrating domain adaptation, edge-aware adversarial training, and noise-pattern disentanglement, DASRGAN achieves state-of-the-art performance across multiple benchmarks and scaling factors. It significantly improves texture fidelity while effectively mitigating structural artifacts. The source code is publicly available.

Abstract

Recent efforts have explored leveraging visible light images to enrich texture details in infrared (IR) super-resolution. However, this direct adaptation approach often becomes a double-edged sword, as it improves texture at the cost of introducing noise and blurring artifacts. To address these challenges, we propose the Target-oriented Domain Adaptation SRGAN (DASRGAN), an innovative framework specifically engineered for robust IR super-resolution model adaptation. DASRGAN operates on the synergy of two key components: 1) Texture-Oriented Adaptation (TOA) to refine texture details meticulously, and 2) Noise-Oriented Adaptation (NOA), dedicated to minimizing noise transfer. Specifically, TOA uniquely integrates a specialized discriminator, incorporating a prior extraction branch, and employs a Sobel-guided adversarial loss to align texture distributions effectively. Concurrently, NOA utilizes a noise adversarial loss to distinctly separate the generative and Gaussian noise pattern distributions during adversarial training. Our extensive experiments confirm DASRGAN's superiority. Comparative analyses against leading methods across multiple benchmarks and upsampling factors reveal that DASRGAN sets new state-of-the-art performance standards. Code are available at url{https://github.com/yongsongH/DASRGAN}.

Problem

• Enhance infrared image super-resolution quality.
• Minimize noise in super-resolution adaptation.
• Refine texture details using specialized discriminators.

Innovation

• Texture-Oriented Adaptation refines details
• Noise-Oriented Adaptation minimizes noise
• DASRGAN integrates specialized discriminator and loss