Study of Colorization and Super-Resolution Efficiency in Image Restoration

this work focuses on the comparative study of the effect of the colorization and super-resolution operations order on the quality of the resulting restored image, with grayscale low-resolution image being the input degraded image. Degraded images assume a complex degradation model. Experiments include direct comparison of the colorization and superresolution operations applied to the data using a number of metrics, as well as retraining a selected superresolution model with a modified, grayscale-augmented degradation model. Results indicate that colorization-first approach is superior to the superresolution-first, and while changes to the degradation model improve the superresolution model performance, the optimal order remains the same.