Summary: IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 10, NO. 1, JANUARY 2001 179
of the median filter, which is used for preliminary processing in most
of the state-of-the-art impulsive noise filters. Our first tests on syn-
thetic and natural images demonstrated that the peak-and-valley filter
and the median filter have comparable performances. The median filter
is slightly better than the peak-and-valley filter at detecting/correcting
noisy pixels, but on the other hand it tends to alter more true pixels.
Nevertheless, even if the amount of true pixel alterations is signifi-
cant (more than 63% for the median filter was recorded), the average
change is rather small (no more than 11 were recorded). In some ap-
plications this degradation of the image, a milt smoothing, could not
readily justify a heavy computational method to discriminate between
noisy and true pixels, especially when dealing with video processing.
The peak-and-valley filter demonstrated to be much faster than the me-
dian filter for all types of images tested.
The peak-and-valley filter represents an interesting replacement for
the median filter in those more sophisticated, and consequently more
performant filtering methods, in order to improve their efficiency.
 A. Beläid and A. Beläid, "Reconnaissance des formes: Méthodes et ap-