Random-valued Impulse Noise Reduction by MST-based Method for Color Image
Takanori Koga, Noriaki Suetake · 2011
Abstract — In this paper, noise reduction performance of a switching vector median filter with a random-valued impulse noise detector for color images is evaluated. As a random-valued impulse noise detector, a method utilizing the minimum spanning tree (MST) is employed. In the switching vector median filter, the impulse noise detector is employed before filtering, and the detection result is used to control whether a pixel should be filtered or not. By applying the method to color images, it is expected that the impulse noise is reduced pointedly while preserving detailed structures such as thin lines, sharp corners, and so on. Through some experiments, for color images, the effectiveness of the combination of the MST-based random-valued impulse noise detector and the switching vector median filter is verified. Particularly, the present method is compared to some powerful switching vector median filters which had been proposed so far from a view point that the impulse noise reduction while preserving edges and details of an image is realized or not.