Rank Replace 8-band (11th, September, 1999)

Description

Rank based replacement in 3x3 convolution, removes noise while preserving edges adjustable individually over 8 tonal ranges. Eight neighbors are sorted and if the center pixel is out of the current rank-range it is replaced by the nearest (either min or max) of the selected rank. All settings of this filter are more gentle than a Median filter.

Selection Dropdown

You can use the tonal range sliders by choosing Very Low Key Image, Low Key Image, Middle  Key Image, High Key Image or Very High Key Image  according to the key of the image that you are editing. Else choose Rank1 ... Rank 5 that will be applied all over the same. 

Rank 1 is the same as the common Min Max replacement filter. It removes spurious isolated noise pixels.

Rank 2 is slightly stronger than 1st Rank but sill preserves edges.

Rank 3 affects to the edges slightly but removes a lot of noise.

Rank 4 is almost as strong as a Median Filter, but does not do as much damage.

Rank 5 replaces with median of the neighbors. It flattens noise nearly as much  as a Median Filter (at radious 1) but preserves edges better, since it only replaces out of rank pixels where median just applies the median everywhere.

All Adjusted selections enables the tonal range sliders. They allow to specify different ranking based on the average RGB value in the 3x3 (less the center pixel) convolution matrix.

Sliders

Sliders are only effective when an  Adjusted mode is selected in the Selection Dropdown (see above).

Tone n/8:  These 8 slider affect to eight perceptually equal ranges, Tone 1/8 is the most dark range and Tone 8/8 is the lightest range. The value of these sliders is the rank 0 to 5. Note that setting a slider to 0 completely disables the operation for that range. (The spline that the Tone x/8 sliders control is smoothed slightly).

Notes

This filter is not very usable in gamma spaces since the two or three lowest sliders (in the black-end of the range) do not notably affect to gamma compensated images. This is not the problem of the filter but is due to the image-gamma. It is possible to write the filter in such way that it 1) removes the gamma compensation, then 2) applies the sharpening and finally 3) re-applies the image-gamma. Since the filter works in 31-bit/color this would not degrade the quality but it makes the filter extremely slow. It is already a little slow because of all the calculation needed.

First thing to do when the filter opens is to Shift-Click the preview + button in order to have the 1:1 zoom in the preview box.  Then drag the image so that a representative area is shown, the dragging is not limited to the preview box boundaries.

Author

Timo Autiokari. Homepage: Accurate Image Manipulation for Desktop Publishing

Usage and Distribution Rights

Free to download and use, distribution prohibited.

Filter Engine

FilterFormula v 1.1, by ATS/Graphics