Image from a Nikkor 10.5mm: a case study for C.A. analysis

Foreword

It is quite difficult to fully comprehend the subtle details of chromatic aberration on the entirety of a "normal" image that is shot with a fisheye lens on a DSLR.

I had recently posted a test report about TCA assessment of several fisheye lenses. The subject is rather "hard to understand" as one reader pointed out. It may also be that I was not clear enough in my presentation. After posting, I subsequently was asked by several readers to show some sample images or illustrations. There was also some digression about the nature of the TCA in relation with the real spectrum of the illumination...

Recall

Following are two graphs that illustrate TCA in the RVB Blue channel of an EOS 5D image shot alternatively with a Tokina zoom fisheye lens and a Nikkor 10.5 mm fisheye lens as typical examples. Both lenses were not originally intended by their designers for use on a full frame DSLR and have been sawed off from their sun shade to extricate the full angle of view possible coverage on the CMOS sensor.

In short: I had demonstrated in the previous article that the lighting type has a strong impact on the magnitude of the Transverse Chromatic Aberration, but this affects mostly the blue channel of the RGB image and not much the red one. This applies similarly on any fisheye lens and not only the Tokina 10-17 mm (Top image) or the Nikkor 10,5 mm (Image just above).


Introduction

Recall: I had tried to simplify the task of image evaluation by designing a special test set-up which main component is a specific cylindrical target. This was done in-house as I could not find such a thing or similar for sale elsewhere;-)

The target is a long piece of paper that looks approximately like this (Cylindrical QTVR 600 kB) when seen from the center of the cylinder that is the precise spatial location of the entrance pupil of the fisheye lens. It is a good approximation of an image of a pure B & W object . You should zoom in at high level to see the spectacular effect of color fringing due to chromatic aberration.

With one "specific" test image sample I shall attempt to show in more detail how the chromatic aberration appears like on the image after de-mosaicing (part of the raw image conversion) and when some manipulations are done and viewed with the help of a huge magnification factor. I shall therefore show the use of a kind of digital powerful microscope to scrutinize the image shot by the Nikkor 10,5 mm at the sweet spot aperture setting (i.e. f/5.6) on my FF camera i.e. the Canon EOS 5D.

As the Canon 5D raw image format is only supported by these recent versions, I propose to use Adobe Photoshop CS2 (or CS3) and ACR3 (or ACR4) from here on.

Part 1: Is there really a single image (of the space object) in each RGB layer?

The sample image (EOS 5D + Nikkor 10.5 - fluorescent light)

The sample image in raw (.CR2) format can be downloaded from there (Beware => 11.2 MB!). I would also recommend to download this .xmp file that can be simultaneously used to get the very same Photoshop Camera Raw settings for conversion from raw format (except the °K setting, every other is kept to ZERO).

Analyzing in minute detail the three RGB layers of the converted sample image.

Convert the image to TIF 16-bits in ACR 3 (or 4) with all settings to zero but temperature ~ 2750°K.

Look at the Red, Green then Blue channel images away from the image center with the magnifier set at 700% for instance. You could possibly also use the "Levels" tools to emphasize the effect

Observations:

Opening the Photoshop "Image > Adjustments" menu you may then use "Levels" or alternatively "Brightness/Contrast" and play in large stokes with the sliders to discover how complex can be the details of every channel! In fact, IMHO you should be able to distinguish numerous images that are there driven out and NOT exactly overlapping with each others.

As expected, the defective aspects of the image grows when going away from the center of the image.

For the sake of comprehensiveness, this is a detail of another and different image (i.e. Tokina 10-17 mm at 14 mm /BLUE channel) that you may compare with the Nikkor reference Sample image: look on the thumbnail image (lower right corner of the main image) for the location of the zoomed-in part.


Part 2: Natural sun-light illumination

The sample image (EOS 5D + Nikkor 10.5 - sunlight)

To see what it looks like with the same target (with B&W motifs) shot under natural sunlight, you may download this file .CR2 = 12.3MB (!) and this .xmp file and then proceed the same way as for the referenced previous image shot under fluorescent light.

Analyzing in minute detail the three RGB layers of the converted sample image

Convert the image to TIF 16-bits in ACR 3 (or 4) with all settings to zero but temperature = 6100°K.

Look at the Red, Green then Blue channel images away from the image center with the magnifier set at 700% for instance. You could possibly also use the "Levels" tools to emphasize the effect.

Observations:

  1. The red channel image shows a very strong asymmetrical aspect of the shading off (optical softness) on the opposite radial edges of the strongly contrasted white motifs against the black background. I can even see a deep black edge on the white checker side closer to the center and a clear shade off white to grey on the side facing the outer part of the image. No need to enhance the contrast to observe this fact.
  2. On the green channel image the shading off on each side of the checkers looks quite symmetrical at first without contrast enhancement. With some strengthening by using the slider of the Level tool, one can see a similar asymmetrical pattern as for the red channel, but reversed!
  3. On the blue channel the asymmetry is visible easily but with a smoother aspect.
Red channel
Green channel
Blue channel
80 degrees from lens axis
80 degrees from lens axis
80 degrees from lens axis
Zoom 700% No contrast enhancement
Zoom 700% Enhanced contrast
Zoom 700% Enhanced contrast

There are no ghost sub-images on any of the RVB channel images.

As expected, the defective aspects of the image grows when going away from the center of the image.


Discussion

(part 1: Artificial fluorescent light)

I believe that the digital manipulation that is described above shows the real nature of the image: each of the converted image RGB channel is the result of the overlay of several individual images that are not stacking perfectly. This gets worse when getting from the center toward the edge of the image circle.
I believe that every individual image in each channel results from a peculiar narrow part of the artificial light spectrum that may come from peaks and valleys in the illumination combined with peaks and valleys in the spectral sensitivity of the sensor or the filter and electronics assembly. I don't know if the raw conversion/interpolation algorithms play additionally a role here.

I had thought for a long while that these multiple ghost sub-images were the cause of the shape (reversing of the gradient) that is observed on the plots shown for instance in the foreword above. I had only experimented with fluorescent light at that time. I now believe this is probably not the case as I have in the mean time be able to make the same experiment under tungsten as well as natural light and I have observed similar trend in all cases.

However and to say the least, this shall most probably have an influence on the performance of the automatic correction of TCA that some developers plan to implement.

(part 2: Natural Sunlight)

The absence of ghost images makes the analysis much easier in this -common- case.

The striking fact is the evident asymmetrical aspect of the gradual grey shading on the opposite radial sides of the white checkers on each of the RGB channel images. This has been a surprise that I personally did not expect to see before experimenting. Furthermore the asymmetry is reversed in some pair of the three channels!

Another fact is the large difference between the width of the shading in the image (i.e. the softness) of any channel and the shift of this image w.r.t. the image of an other channel (this shift is the real TCA). I had, like many others, abusively called "chromatic aberration" the obvious color fringe that anyone can observe on the RVB fisheye image before CA correction. I was wrong: shifting adequately both the red and blue channel images by just and only a few pixels shall (hopefully) transforms the much larger RVB image apparent "color fringe" into a residual shading off the edge. This residual defect cannot be corrected and is the actual "softness" of the lens. The larger the apparent "chromatic aberration" (aka color fringing) is before TCA correction, the softer is the lens even after TCA correction.

BTW if all this is more obviously viewed here under natural light, it must eventually be true for any type of lighting, including of course the classical tungsten bulb light.

Conclusion

From this type (btw that encompasses also the Tokina 10-17 mm zoom at least) of fisheye lenses, the images shall sometimes exhibit blue (more or less tinted with magenta) diehard fringes of very contrast bright edges.

I have experimentally verified that with any types of lighting, the asymmetrical aspect of the shading (optical softness) on the edges of the strongly contrasted white motifs against the black background (especially on the red channel but not only) cannot be compensated by simple geometrical transformation such as done by TCA correction tools. Stacking R, V and B such images yields blue or magenta (depending on light conditions and image pre-processing) fringing of these high contrast edges. IMHO this effect cannot be fully corrected.

This defect shall be noticeable on specific occasions (e.g. branches of trees against a hard backlighted and overcast sky during a winter afternoon). It should be present only on one edge side of the branches i.e. the edge that radially faces the image center or the edge that is radially opposing it, not both.

One important discovery is that this exists even when the brighter side of the concerned edge is not overblown and over exposed at all. Thus this should not be confused with the famous other color fringing (again!) resulting from hard back light that can produce a blue, violet or purple fringe if some sensitive CCD or CMOS sensor elements are saturated and overflow the neighbors with electrons... but it shall look almost exactly like it as there is no complementary color (yellow or cyan) on the opposite side of the bright object in the RVB image.


Part 2: Isn't TCA function of aperture f-stop setting?

On the above chart, all the f-stop plots for chromatic aberration on the Red RGB channel have been composited.

On the above chart, all the f-stop plots for chromatic aberration on the Blue RGB channel have been composited.

Conclusion

Taking into account test and computation uncertainties, there seems to be negligible variation across the f-setting range.

--> TCA doesn't depend on aperture f-stop setting .


Michel Thoby

2 July 2007

Rev: 5 July 2007