Fixing bad purple fringing

divamum

I know that once upon a time there was an action that fixed this, but last time I followed links for it, it was no more...

Is there a way of recovering those areas? I have a shot with some REALLY bad purple edges (85mm 1.8 - nuff said!) and am trying to figure out the best way of getting rid of minimizing them.

Thanks in advance!

pathfinder

I posted this link on Nov 21, Diva - it is the best technique I have seen, and you can easily convert it to an Action of your own.

http://vimeo.com/3069865

This is a link to an earlier post of mine in that same thread - http://www.dgrin.com/showpost.php?p=1505105&postcount=5

pathfinder

You may want Defringe -> All Edges in your Raw converter as well. That will help quite a bit.

divamum

THANK YOU! (and sorry I missed earlier post with this info - oops!). I'll play with these as I work through my processing on the affected shots.

I did use the edge defringer in LR, but it didn't make any dent in it. These are BAD ones (and the 85 1.8 is notorious for it- haven't used it in a while and had totally forgotten just how bad )

pathfinder

I own, and like the EOS 85 f1.8 too. It is a very poor lens for IR work, but I have not really been troubled by purple fringing. I just shot a bunch of family kids, with just the 85 f1.8, and just finished editing them. All at f2.2 ( or very close ) and ISO 1600 with my 7D. Many strongly backlit from outside facing windows. Some frames have some R/G CA that needed correcting, but real purple fringing I did not notice.

Is this true of all EOS 85 f1.8s, or just some of them?

Sam

Great tutorial!

I will need to try this!!

Sam

divamum

All of them are subject to it afaik. I've now had two copies - bought it, sold it for a 100 f2 while I saved for a 135, got the 135L and swapped the 100 out for another 85 1.8 (there was a focal-length method in my madness! ) - both were afflicted, and if you read reviews/user reports of it it's almost always something that's mentioned. Particularly bad on strong highlight edges.

pathfinder

Maybe I'm just iucky

I wonder if it is more of a problem on a FF camera, or a crop body box....

divamum

Dunno, but "85 1.8 purple" pulled up several pages of google hits

http://www.google.com/#sclient=psy&hl=en&q=85+1.8+purple&aq=f&aqi=&aql=&oq=&gs_rfai=&pbx=1&fp=3f2d83e828bdfbc

Do you tend to shoot stopped down? If memory serves, it's worst at wide apertures...

pathfinder

As I said, I was shooting at f2.2 all afternoon last Saturday - They are private files, but I will PM you the details so you can look at them.

MarkR

Do you have a lens profile you are applying in LR3? I seem to remember seeing some PF going away once I applied a custom profile-- seemed to get fixed with the CA and distortion... I could be wrong tho. Also, are you viewing at 100%? I've seen what looked like PF go away once zoomed to 100%.

divamum

Thanks Mark - it's only really visible when magnified since the area affected is a small part of the overall frame. Tried the lens profile, but sill no luck.

Currently doing it the hard way: sampling sections on the blue/magenta channels and doing layer masks for each colour affected - sadly, those colours are components of other areas of skin, so I can't just reduce them and walk away. Phooey!

ETA:

Here are deep crops of the shots (just about 100%):



And what happens when I start messing with magenta swatches (pulled all the way down to the show how other areas of the shot are using the same colours, but it's noticeable even if I don't reduce saturation on that colour as much as this)

pathfinder

Are you sure you have used the correct lens profile for lens correction in LR3 first, Diva? If the profile gave you no joy, I would then try manual correction of CA with the R/G slider with Defringe all borders turned on.

There is some real green bordering in the background, that I would not call PF, but CA, just to the left of the magenta margin of his hat. I would make a strong effort to correct this with CA tools first, I think.


If you have no luck correcting this in software, I think if it were my lens, I would want Canon Factory service to look at it carefully. You saw what my 85 f1.8 files look like on the same camera body, 7D. My files do not have this type of issue, and I don't think you should either.

This is pretty significant CA, or PF, or whatever. Even my Tamron 28-300 has no where near that degree of CA.

divamum

Thanks for the input, PF, as always

As mentioned above, I've had two copies of this lens, and they both did it (the other one was rather worse, in fact) - different camera bodies, too, so not that either. If I hadn't seen so many comments about/ reviews mentioning CA/PF on this particular lens I'd be worried about it being a lemon but, as it stands, I think it's just the nature of the beast. Not sure if you got lucky with yours, or if your exposure technique is so good that you manage to circumvent the issue before it's a problem (my guess is on the latter, for what it's worth )

Bottom line is that I blew the highlights on this shot, drat it, which is entirely my fault - if I didn't like it so much, I'd just bin it, but it's a very nice one of the set that I'm hoping to preserve since other than this, the flaws don't really detract in a significant way. Fortunately I can fix this "the hard way" (thanks for those HSL details and techniques) ... it's just more work!!!

MarkR

Divamum -- another possibility might be to create your own lens profile using the Adobe tool. That *should* give you better results than using the canned profile. And it's not that hard. Even I could do it.

pathfinder

Or just start with the standard profile, and start sliding the CA slider, and then switch to Manual and Defringe all edges.

divamum

I'm not at home right now to try again/check, but I was perplexed by the CA tools in general when I started playing with them last night, because I didn't notice ANY changes. I know in the past when I've tweaked with those sliders Ive been able to see things alter SOMEWHERE in the shot, but nothing at all anywhere at all this time. To the point I'm wondering if I"m doing something wrong or have something set to prevent it working ...

pathfinder

You have to have the image enlarged to 300% or so to really see the changes real time as you perform the CA corrections. Start with Defringe All Edges, and then if need be, the R/G slider, and then if visible the Y/B slider if you are doing Manual ( not lens profile ) corrections.

divamum

Ah, ok - that would be why. I was maximum 100%. I'll check it out again when I get home tonight

TheSuede

DivaMum, to keep in mind - "what is PF, really?"

This might help you combat it in PP.

Most of the time it is an avalanche effect of longitudinal CA - wavelengths are not focused at the same distance. This is usually only visible as slight discolourations of high-contrast edges, but if you combine this effect with SEVERE overexposure (think branches against blown sky, the highlights on the guy in your sample picture's nose, chin and right cheek, and so on) it can get really ugly. In simple words, the individual channels have different amounts of haze and defocus.

So, what happens when you translate this into an RGB picture?
-Look at the individual channels! (full colour, R, G, B channels)



(1) How do we correct this (start on a copy layer!)?

(2) Oversaturation - easy, do a channel mix R= 70R,30G,0B; G=20R,60G,20B; B=0R,70B,30G
this mixes a bit of green into the blue and reds channels, and removes oversaturation from the green channel.

(3) Remaining halos in red and blue - just as easy, do a simple UnsharpMask. In this shot I used r2.5, 30%. ONLY on the individual R and B channels!

Ok, it looks grim, but what we have done sofar removes 90% of the negative sides of the PF.
Now - in what regions of the picture is the effect most prominent? In areas right next to blown areas (white). If the area next to the blown area is dark, then the effect is stronger. What could be a good masking routine for this?

-A simple highpass fits this description perfectly.

(4) Make a layer copy of the "corrected" layer, and try some HP radiuses on it. Look for the amount of dark the filter produces, you want that dark area to encompass the PF. r12-15 seems about right for the sample picture.

(5) We want the now dark regions of the HP-filtered image to be the transparent parts of the mask. There's several quick ways to do this, I usually just invert [ctrl-I] and then

(6) use curves to make the mask black outside the regions that I want and "just enough" of the bright areas bright. Load the layer as a selection [ctrl-alt-2], select layer underneath, create mask. Delete the HP-filtered layer, it is no longer needed.

(with HP-filter - inverted - with some curves, ready to "load as selection")


Done.
You might want to run a gaussian blur on the mask to extent it a bit into the masked areas - this depends.

What started as this:

Now looks like this:


Not perfect, and certainly not optimized in any way - but easy to apply (an action containing the steps above on my 24MP images takes less than a second) - but you can still play around further with the mask if you absolutely want to.

divamum

Wow!

I'm not sure I entirely understand what you did - and I'm going to have go through that step by step myself before I do, I suspect - but WOW! THANK YOU for taking the time to post that!!!!

TheSuede

Just give me a holler if anything was unclear, I kept the text as short as possible.

I just find the "colour type selection" only variety of PF editing flows to be a very, very blunt tool. They often induce as many problems as they remove. You really need to target (mask) JUST the PF if you want good, reliable results in an automated workflow or action.

divamum

I haven't looked at it again - I'll definitely get back to you if I need further help! I left the shot alone, thoug - if they order the shot I'll revisit this thread and fix it up (I the SM proofing days - it's a godsend with something like this!). I really appreciate the tutorial and help!