The minimal blur kernel: x 12
nel is equivalent to a truncated Gaussian smoothing with a radius of 0.25 pixels. It passes 0.25 of the central pixel.
The "hole in the donut" kernel: ^ x creates a new central pixel that totally ignores the value of the central pixel in the old one. If this filter is applied to an extremely noisy image, it will eliminate noise spikes; or more precisely, it will spread each noise spike into a fairly inconspicuous ring of eight pixels. Its pass-through rating is zero.
Larger smoothing kernels work in a similar fashion, but they operate over a larger neighborhood; so the potential for the desired smoothing and undesired loss of detail are both greater.
Was this article helpful?