Image Noise filtration By Random Point Processes

A new method for identifying pixels as "corrupted" by noise is proposed in this paper for effectively denoising corrupted images. Unlike conventional filtration approaches that apply the filtration operation to each pixel unconditionally or by implementing a noise detection mechanism before filtration we propose to identify noise pixels by their position. As a model for noise pixels placement, we consider random or regular point patterns. To distinguish such pixels placements we form point patterns for image intensity from predefined range and apply a Clark- Evans test to identify the right ones. Next, the conventional filtration approach is applied to each pixel of intensities that form random or regular patterns. This approach allowed to increase a PNSR value and to reduce DSSIM one.