Puzzling CA issue

Discussion in 'Image Processing' started by Underwater, Apr 7, 2015.

  1. Underwater

    Underwater Mu-43 Veteran

    Jun 1, 2014
    Eugene, Oregon
    I'm fairly new to LR, and I'll admit I have a lot to learn as far as post processing. Those caveats lead to my question- I've noticed some horrible CA with shots from my E-M1 when I drop the exposure and highlights in certain underwater scenes shooting up towards the sun. I've noticed it with the 9-18 behind a dome, and with the 12-50 behind a flat port- both with different shutter speeds. The example below was shooting up through some pilings at the sun with the 9-18, and I'm having difficulty removing it in post. I'm guessing a lot of this is from trying to recover too much from the (intentionally) blown highlights, but I don't remember it ever being this bad with my E-PL2- is there something unique about this problem with the E-M1? The bummer is that I also notice it at 1/320- that's the max sync speed, and one of the reasons I wanted to switch bodies in the first place. Any thoughts?

    Attached Files:

  2. pellicle

    pellicle Mu-43 Hall of Famer

    Feb 10, 2010
    Southport, OzTrailEYa
    looks that way to me ... its not CA if you ask me

    less agressive recovery? Accept that (without hdr) that there is a limited range and once your exposure moves into shadows that highlight areas will go 'radioactive'

    some systems do some good recovery blending, have you done any colour shifts or corrections of light balance which may cause that shift into the cyan? If yes, then probably that's 'screwed' with your (fade to white) recovery
  3. David A

    David A Mu-43 All-Pro

    Sep 30, 2011
    Brisbane, Australia
    Explanation? All bets are off. It would be much easier to diagnose if you took it above water but you were below so at the very least the following things are occurring:

    1- water has a different refractive index to air so light rays are going to be bent in different ways in water.

    2- water absorbs light progressively as you go deeper. I have no idea how deep you were but that could have an effect on the colour of any chromatic aberration.

    Plus I'm not certain that I can see any chromatic aberration. What I can see is some flare and what looks like some diffraction of the light as it passes around the piling. Chromatic aberration shows either as rainbow fringes around the edges of objects or as green or purple fringing depending on whether it's lateral or longitudinal chromatic aberration. I can't see any of those effects in the image you posted but there may be something of that sort visible at greater magnification on your screen. The colour banding I see in the blown highlight area looks like it may be due to a change from clipping all 3 colour channels in the white blown areas to only clipping 1 o2 2 channels in the slightly darker areas which are showing a trace of colour, and if so the colour will be dependent on which channel or channels aren't clipping.

    As for differences between the E-PL2 and the E-M1 which may make a difference, the E-PL2 has an anti-aliasing filter and the E-M1 doesn't, plus they are different generation sensors. I remember being impressed by the difference in the sensors when I moved from an E-P3 to an E-M5 because the E-M5 had a greater dynamic range and a lot of that gain in dynamic range seemed to be in the highlights. I think my E-M1 gained a bit more in the highlights over the E-M5 but doesn't seem to be quite as good with low light and deep shadows where the E-M5 seems to have a slight edge. In other words, there are certainly some sensor differences which could be making a difference.
  4. dwig

    dwig Mu-43 Top Veteran

    Jun 26, 2010
    Key West FL
    +1, plus ...

    The highly recovered area only spans a very narrow range of the original images brightness levels. When corrected like this you expand the difference between the brightness steps but do not create new levels in between. The result is the "banding" seen in this image.
  5. Underwater

    Underwater Mu-43 Veteran

    Jun 1, 2014
    Eugene, Oregon
    Thanks for the info, everyone- it's always nice to learn something :) I kind of suspected it was an issue unique to underwater, and since it was purple, I immediately thought CA. The shot was taken at about 10-15 feet below the surface, and I didn't touch WB in post. I think diffraction coupled with Dwig's reference to banding explains it. The good news is that I'll get to experiment some more this weekend.
  6. pellicle

    pellicle Mu-43 Hall of Famer

    Feb 10, 2010
    Southport, OzTrailEYa
    nobody does this by hand anymore (except dingbats like me) but no matter if its you or the automated software, highlight recovery still works mathematically exactly the same way. This is an old tutorial on how the basics of highlight recovery works on dcraw (which some processing systems are still based, such as Rawtherapee) and his demonstrations show something quite like what you have


    obviously this example is a shirt blown out not the light above. The ideas of banding are in my view a red herring because banding is most commonly associated with posterisation (nb weren't enough levels to allow streching of the tonal range) but this applies mostly to shadows as this is where there is not enough data to support it. This is because of the combination of sensors register data in a linear way and we prefer to represent it in a logarithmic way. For instance:

    you can see in the above figure how a small bit range ends up being streched, yet the high end is compressed ...

    Why the problems come in is that 1) the sensors do not run out of data at equal points (causing weird colour shifts in the clipping area) and 2) some colour correction is applied even if you are using "fixed" white balance.

    If you are interested, there is way more exploration of this in my old blog post here

    But you can see in this analysis of a RAW file from my G1 that for a given "light value" the green started recording way before the other channels were registering ...


    and ended while they were still going (note the clustering of data in the red channel at pixel levels 3988 (which in binary is about 12 bits of data ... which may help you to get the DxO metrics more).

    Best Wishes
  1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.