An efficient operator-splitting method for noise removal in images

In this work, noise removal in digital images is investigated. The importance of this problem lies in the fact that removal of noise is a necessary pre-processing step for other image processing tasks such as edge detection, image segmentation, image compression, classification problems, image registration etc. A number of different approaches have been proposed in the literature. In this work, a non-linear PDE-based algorithm is developed based on the ideas proposed by Lysaker, Osher and Tai [IEEE Trans. Image Process., 13 (2004), 1345-1357] . This algorithm consists of two steps: flow field smoothing of the normal vectors, followed by image reconstruction. We propose a finite-difference based additive operator-splitting method that allows for much larger time-steps. This results in an efficient method for noise-removal that is shown to have good visual results. The energy is studied as an objective measure of the algorithm performance.