Denoising images corrupted with impulse, Gaussian, or a mixture of impulse and Gaussian noise

Abstract

In this paper, a cascade of stages is used to denoise images corrupted with Gaussian noise, impulse noise or a mixture of the two. The proposed method is based on removing the abnormal values in the initial stage to help remove the noisy small components in the next subsequent stages. The initial stage is based on an intensity difference technique, while the subsequent stages are based on principle component analysis. Most of the noisy small components are removed in the first subsequent stage and then the smaller ones, which represent the remaining noise, are removed in the next stages. At the end, a new restored version is attained and compared with the results of well-known methods. Simulation experiments illustrate that the proposed method is very fast and delivers superior performance either in terms of Peak Signal to Noise Ratio (PSNR) or visual image quality, particularly in restoring images corrupted by a combination of Gaussian and impulse noise.