Enhancing Line Sharpness in Blind Image Super-Resolution using Diffusion Models Color Swapping

Real-world images especially in satellite images, often have low-resolution (LR) images due to being captured from long distances. This makes it difficult to perform computer tasks such as path detection and object recognition. One of the solutions is to implement Image Super-Resolution (ISR) to upscale the LR images into high-resolution (HR) images. Recent method of ISR is diffusion model that has shown promising in generating high quality of images color; however, it lacks of line sharpness due to residual shifting limitation. This research address the problem by modifying the parameter of the diffusion model to create HR image that has better line sharpness, unfortunately the color is too sensitive, thus this study integrates those two HR images by color swapping based on YCbCr color space. Experimental result on a road scene CH6CN satellite image dataset shows that the proposed model performs better than previous diffusion model with improvement of 1.10% and 1.14% on NIQE and NRQM.