Basic algorithm:
If center > TC and mean
(sum) of 20 neighbors > TN, then, set the pixel '1'
TC = threshold for center, (get from the pooled histogram
of normalized deviations, out of empty training sequence, at the point
of the false alarm rate).
TN = threshold for neighbors, (get from the pooled
histogram of mean (sum) deviations of neighbors where the center = 1 at
the stage of TC, out of empty training sequence, at the point of
the false alarm rate).
Tother =
threshold for original deviations, which is comparable with the
support filter results.