Comparative Study of Image Degradation and Restoration Techniques

This paper implements airy disk smoothing, Poisson noise, Gaussian smoothing, Hanser’s phase term, and Zernike polynomial phase term as degradation techniques on images from the DIV2K dataset. These actions allows the generation of gray-scale degraded images to study the performance of the inverse filter, Wiener filter, and the Richardson-Lucy algorithm as image restoration techniques. Our experiments are conducted on two representative tasks: (i) intense image degradation, and (ii) image restoration from the degraded images. To measure the image degradation and the image approximation to the original image, this paper uses four similarity metrics: global dimensionless relative error of synthesis (ERGAS), mean squared error (MSE), spectral angle mapper (SAM), and visual information fidelity (VIFP). These similarity metrics determine which restoration technique can estimate the original image in more precisely, and enable the analysis of the required conditions for the estimation.