Speeding Up Powerful State-of-the-Art Restoration Methods with Modern Graphics Processors

Abstract

One important aspect of digital image processing is the removal of glitches from the measured data, particularly from observations of physical phenomenons. We propose an approach which realises valid results that have nearly no restoration artefacts. It is based on a further developed state of the art regularisation principle - the restriction of the degrees of freedom of the solution-describing model. The key idea is to use parameterised image elements instead of single pixels which are determined jointly in a bayesian estimation. However, the long duration of a restoration using such an approach is a problem in many applications. This article presents a technique how to speed up this method and reduce the runtime using the example of restoration of kelvin probe force microscopy-data.