In this paper the problem of reducing the noise in a digital image preserving information about edges and corners is addressed. To this aim a new technique is described, based on the USAN principle. This principle at each image point swaps the pixel value with a new value, which is more representative of the region where the point lies. This technique is a practical alternative for efficiency and quality of the produced image to the offset-filtering technique proposed by Fischl and Schwartz in [4]. A set of experimental results on artificial and natural noisy images highlights that the new technique has a greater noise reduction capability and preserves edges and corners from smoothing better than other published methods.
展开▼