Accurate filter for removing impulse noise from one- or two-dimensional signals

This paper is concerned with the suppression of noise from one- or two-dimensional signals. Commonly, these will arise as acoustic waveforms or images from a variety of sources. The aim of this work is to ensure that the resulting signals are as distortion free as possible. Preceding work using the median filter has shown that it can result in edges being shifted by noise or (in two dimensions) as a result of the curvature of object boundaries. Here the concern is with the effect of impulses in shifting object boundaries. A filter has been devised with the aim of removing such impulses while maintaining the remainder of the data unchanged. Hence, later application of a median or other filter should produce no specific impulse-induced distortion. Tests confirm that, at least for isolated impulse noise spikes, this is achieved. The results are shown to be of particular value in image interpretation using Hough transforms.