Kendall Helmstetter Gelner wrote:
The railings clearly show in front of the left (bent) leg. The area around the right leg is too busy to make out the railing even against the black pants.
They still fade out in the bent leg.
If it's CA at work, what lens gives you flesh colored CA?
If you have a blue object on a red background as here, the blue will run into the red and vice versa.
But the same happens again skin too, which again is not really red at work.
Can you do the resizing again using nearest neightbour interpolation
They appear to be about 2-3 pixels wide.
I said the resolution limit of the "sensor". Bayer sensors have the lowest resolution for red / blue, that's well established in colour resolution charts. Your images are consistent with my "supposition".
I'm not sure, since the railings are a little wider than just a pixel.. but for sure the lower red/blue resolution is playing a part there, I would say for sure in the case of the box and somewhat less in the case of the runners legs.
Indeed there is some color bleed through, but not an outright vanishing as with the railings against legs - instead the falloff is more natural.
The red lines also did not vanish in the equivalent D60 image.
Ahh, but they do show dropouts in the middle, randomly, which I would argue is actually less preferable to a simple dropoff because it's more notable.
Interesting that you have not brought out the equivalent Foveon image for that railing scene to extoll the virtues of Foveon instead of just bashing Bayer. How did Foveon fare in that scene now, really?
I would love to - but despite the M9 and D3X images coming from camera reviews on two totally different sites (one is DPReview, the other Imaging Resource) neither of them bothered to use that same scene with any Foveon based camera!
The next time I'm in London I'll seek to replicate the shot at different focal lengths to get near a single pixel width for the railings. Or, if I can find any blue railings locally I may try some tests. Blue railings are harder to find than you would think.
Well that's what happens when you shoot a DP-1 and overexpose reds, later cameras did not have this issue. Also the JPG conversion (in camera JPG I think used for the reviews) may have altered the tone a bit.
DP1, D60, etc. They were all shot in jpeg.
Right, but everyone knows the bayer cameras have a lot more advanced JPG processing engines, certainly moreso than the DP-1 at the time that was taken.
I was hoping they would have an X3F fie with the review for that image but I could not find one. Im looking to produce a red/blue color chart to see if in raw the effect would be the same or as pronounced.
It's aliasing, clearer if you look at the small circle. Happens to Foveon too on the similar patterns if the sensor resolution limit is exceeded, which in this case it's not.
I'm not talking so much about the spatial aliasing causing the odd patterns, more like single pixel dips into the larger red bands making them look kind of choppy.
So we have established that
- when the resolution of the sensor (or the lens) is exceeded, colour errors occur in colour line pairs regardless of sensor technology (simple in the Foveon case: if you have red and blue lines denser than the pixel density, you will have both red and blue on every pixel.
I don't think that's what is happening in either image though, since you see the purple show up in the foveon image earlier than the single pixel case.
- the colour resolution limit of Bayer for some colour pairs (especially red / blue) are lower than half the Foveon pixels, so red / blue line pairs will start to run together earlier than expected, especially when you zoom to 200% (when you really ought to be zooming down to 70.71% to halve the pixel count to make a "Foveon equivalent")
I didn't want to go off even powers of magnification so as not to introduce artifacts, which is actually also why I chose bicubic for upsampling as I thought it would be most representative.
- on the other hand, it will resolve more than half the Foveon pixels' worth of line pairs with less colour, and resolve much more than half for black and white line pairs.
- therefore, on average Bayer resolves the equivalent of 50% the amount of Foveon pixels. Of course, it may resolve rather less than that for a person with a fixation on red and blue line pairs
Sounds about right.
- there is nothing special to Bayer about the loss of detail or colour shift around colour line pairs that isn't compensated for overall by extra resolution in more favourably coloured areas. As per point 1, Foveon sensors also colour shift under similar conditions, it just has a higher threshold (but a lower threshold for monochrome lines)
Again, I'm not sure the railings are small enough for that to be quite what is happening. But it's an interesting case to examine.
--
---> Kendall
http://InsideAperture.com
http://www.pbase.com/kgelner
http://www.pbase.com/sigmadslr/user_home 
