#280 Purple seams aournd highlights not cleaned up
I wanted to develop a CR2 image with UFRaw. It was taken with
the notorious EF 50mm/1.4 and contained blown highlights. Therefore,
I dailed in
- AHD interpolation,
- LCH highlight restoration (but no exposure compensation),
- camera white balance, and
- color smoothing enabled
into the latest CVS version of UFRaw. And BTW, I already checked that
color_smooth() in "dcraw_api.cc" really gets called with "passes = 3"
in my case.
To my surprise the result shows purple seams around some of the
highlights. If I however develop the image with the current version
of dcraw (revision: 1.442) as contained in the UFRaw CVS project with
the options "-m 0 -q 3 -H 0 -w", there are significantly _less_ purple
fringes than with UFRaw (same repository state). Developing with "-m
3 -q 3 -H 0 -w" makes all purple artifacts go away and IMO perfectly
cleans up the image -- as expected.
This is strange. Highlight restoration is not suppose to fix purple fringes due to lens chromatic aberrations (CA). UFRaw does have an option to fix CA in the lens correction tab. Did you try it?
The seams are not caused by CA: (1) They occur in the center of the
image, too. (2) Changing k_r or k_b only worsens the problem. -- I
tried it as you recommended before. (3) A purple seam means the blue
or red channels "overshoot" with respect to the green channel; there are
no bluish or reddish seams.
To me it looks like a perfect application of the median-filtering
technique, i.e., the "-m#" option of DCRaw or Section 3.3 in the
Hirakawa and Parks paper on the AHD-algorithm.
I don't understand, why UFRaw does not get rid of the purple fringes,
despite calling its median filter routine. RawTherapee and DarkTable
let me specify the number of applications of their median filters just
like DCRaw's "-m", and both get rid of the purple fringes exactly that
way.
Is the problem still present in UFRaw 0.19.1? If so, please make a test file available for download.
Regards,
Niels Kristian
CVS. THX for your continued interest, however you could reassign
minimum priority to this issue, for I have switched from UFRaw to
DarkTable (also because of this problem, as DarkTable cleans up
almost-completely blown highlights far better).
WRT the original raw image: I sincerely distrust all file
sharing/stealing sites. So if you want the original data (~31MB)
please tell me a semi-public FTP host or something along that
line, where I can drop the picture.
You can send it to me through www.yousendit.com. My mail address is nkbj(at)users.sourceforge.net.
Regards,
Niels Kristian
Thanks for the sample file. I can see the purple rims around the highlights in the water. I will look into it, when I find time, but it will be low priority.
Regards,
Niels Kristian
Hi,
is this issue still on the radar? I'm suffering the same problems when developing my RAWs with UFRaw. No CA (I'm correcting for that with lensfun), and also for lenses of superior quality (Olympus MFT 75mm), and only at/around saturated areas. Another sample available at http://www.royac.iac.es/~pit/p1080237.rw2 (Panasnic G5 raw image).
I'm using the CVS version of UFRaw, currently as of January 11, 2014
Thanks for your work!