Infrared Image Super-Resolution with Multi-Domain Attention and Spatial-Frequency Processing

Chun-An Lin, Yu-Chia Chang, Tsung-Jung Liu

Abstract: Infrared (IR) imaging is crucial for night vision and environmental monitoring. Our research introduces the MDASRN network, which integrates Multi-Domain Attention with Spatial-Frequency Enhancement to improve IR image resolution efficiently. Utilizing a Multi-Domain Attention Module and a Spatial Frequency Block with FFT and IFFT, MDASRN harmonizes various attention mechanisms and merges spatial with frequency domains to refine image textures and edges. Tested on the DIV2K dataset, MDASRN demonstrates superior PSNR and SSIM metrics, highlighting its capability to enhance image quality in computationally constrained settings.