| Title |
Separated Learning Model Between Colorization and Super-resolution for Low-Resolution Gray-scale Image Restoration |
| Authors |
권순용(SoonYong Gwon) ; 서기성(Kisung Seo) |
| DOI |
https://doi.org/10.5370/KIEE.2023.72.3.434 |
| Keywords |
Deep learning; Colorization; Super-resolution; Image restoration; Decoupled Loss |
| Abstract |
Restoring low-resolution gray images to high-resolution color images is a challenging task known as the ill-posed problem with no fixed answer. In addition to the traditional image processing techniques, deep learning methods have recently been attempted, but it is still very difficult to restore images naturally. Various and effective approaches have been developed for each colorization and enhancing resolution of images, but simply combining the two techniques results in accumulation of errors. To solve the above problem, we propose a network that separates the loss function between colorization and super-resolution with adding a super-resolution model in parallel to maintain the performance of super-resolution. The performance of the proposed method on the DIV2K, ImageNet-1k validation dataset was compared to others via PSNR, SSIM, and FID metrics. Experimental results show that our method outperforms existing methods. |