Purple Fringing

[tomczak]

Thanks Den, I'll certainly report on how this works as soon as I have a chance to try it.

I just realized that most (all?) the techniques above rely on masking the purple fringes and than globally tweaking their hue and brightness so that they don't stand out so much anymore. I figured that perhaps it could work better if the purple fringe pixels were replaced (filled in) by surrounding pixels or their tones (akin to median, but with larger kernel, a magic brush of sorts). I haven't been quite successful, but there is promise - with the purple fringe pixels are masked as before try:

1) Gaussian blur with a larger radius
2) Advanced Sharpen speck removal tab

I also tried high/low pass composite, but I got lost and did not accomplish much.

In paint tool, there is an option to replace the Hue/Saturation only. Maybe that's a path to take, but that would probably have to be done in the clone tool - it's doable, but I think I would have to separate channels first. Any other ideas or hope for automation? I think purple fringing is prevalent enough to spent some time on figuring out how to do it right and with the least effort. Cheers.

[tomczak]

Would anyone have a clear idea as to either what causes the purple fringing or/and what technically causes the purple fringing to occur in images (e.g. what happens to channels that makes the purple fringing show up). Maybe that's a good way to figure out how to fight them, similarly to chromatic aberration transformation that goes after the mechanism in order to fix it.

[Dieter Mayr]

Maciej

The reason of PF seems not to be finally determinated, but i found a explanation that sounds plausible to me:
http://www.dpreview.com/learn/?/Glossar ... ion_01.htm
It says is a effect of the microlenses in front of the sensor, which would explain why different camers with the same lens can show dfferent effects of PF.

Hope it helps

[tomczak]

I'm not quite done experimenting, but I was hoping that others could try it too: using the Remap to automate PF removal. It seems to me that it is possible to set a colour pair(s) and adjust the radii and tightness so that the purple fringes are greatly reduced without obliterating anything else, and without having to set the explicit mask. If this works, the procedure can be automated. Den's generic calibration images could make choosing the colour pairs even faster and more convenient.

I also noticed something else while doing it: while PF are most visible at the contrasty edges, it seems that there is often more subtle purplish casts elsewhere in the same images. Neutralizing the narrow purple range globally (without mask) seem to improve the other areas as well.

[tomczak]

I have a question about the Remap - perhaps someone more experienced could answer it off-hand:

1) What exactly does 'Tightness' regulate - the Chroma and Brightness radii already control how close the hue, saturation and brightness of the search colour in the pair must be to qualify for remapping. Would 'Tightness' be an additional multiplier that globally relaxes all the chroma and brightness radii for all colour pairs?

2) Would it make sense to 'peg' colours, similarly to the Colour Correction transformation, if I want to exclude them? If so, should chroma and brightness radii be set to 0 for those pegged colours?

Thanks.

[den]

MaciejT...... . .Others..

...perhaps this web positing will be of some help re Remap and PF...
http://www.ncplus.net/~birchbay/misc/PF ... emap01.htm

[tomczak]

I tried Remap on Den's test S=50% hexagon, and indeed the replace colour acquires a greenish tint (H=25%) while either increasing the Chroma Radius or decreasing Tightness. The Hue stays almost constant in any combination of Chroma Radius and Tightness settings, whereas V and S increase with increasing Chroma Radius and, especially, with decreasing tightness.

I thought that the Radii and Tightness affected the search colour range only, not the replacement colour. How does it work?

[tomczak]

I resolved the 'pegging' issue I asked about two posts ago: the colour can be pegged; the greater the Radii, the wider the range of tones that will be protected from remapping by other colour pairs.

[den]

The following is a suggested and updated PurpleFringe Hue Reference Image and updated MaskTool-ColorRange pre-set HSV white/black slider positions for finding a feathered Hue range of 66.7% [Blue] to 83.3% [Magenta] distributed throughout an image....

The revised reference image for download:

aPurpleFringe_25-75Sats_0-100Values.jpg (24.27 KiB) Viewed 5070 times

To use the Reference Image: (1) click on it and open MaskTool-ColorRange; (2) left mouse click in the center of the Reference Image under 75; (3) click on Contract; (4) left mouse click hold and drag from the Reference Image center [75] left to 70.9, then right to 79.2, then back to 75 and release hold; (5) move the H [black], S, and V sliders as illustrated; and (6) close the MaskTool. The ColorRange will remain unchanged during a single PWP5 session unless revised later in the session.

The revised HSV ColorRange slider positions:

PFColorRangeHues_masked.jpg (34.66 KiB) Viewed 5070 times

Also illustrated is the HSV ColorRange "'lower-left' Apply - Add-ed" to a 50%Saturation-50%Value color hexagon on a 50% gray background.

With the S and V white/black sliders at their extreme left and right positions, all saturations and values with a feathered hue range of approximately 71+% to 79+% will be masked... Change the H white/black sliders to include more Blue and/or Magenta as needed for a specific image.

Ask if there are questions....