Single Infrared Image Super-Resolution with Lightweight Self-corrected Attention Network

We propose a deep-learning-based single image super-resolution (SISR) method for the infrared imaging system. We construct a self-corrected attention network (SCANet) to reconstruct the high-resolution (HR) infrared image of a target from the low-resolution (LR) one. Specifically, we design a self-corrected attention block (SCAB) that incorporates upsampling and downsampling operation with attention module in recursive and feedback manners. With SCAB, we train an end-to-end network with infrared images to achieve the reduction of parameters and computational cost. The effectiveness of the proposed method is verified by extensive experiments. The results demonstrate that the SCANet can realize single infrared image super-resolution with multiple upscaling factors (e.g., x 2, x 3 and x 4).