Restoration of Noisy Multispectral Images with a Geodetic Distance Filter

Abstract

A number of algorithms for image filtering based on the geodetic distance kernel have been proposed recently. When applied to noisy images, some of them yield restoration results that are comparable to those provided by the best modern algorithms. An algorithm of this class that improves the restoration result for multispectral images is proposed. The algorithm uses a convolution kernel based on the geodetic distance and offers a number of advantages, since it allows for recursive computation and, consequently, rapid image processing. The quality of image restoration by the proposed algorithm increases as a function of the number of image channels, which is important for restoration of multispectral images. Several criteria are used to estimate the quality of image restoration.