Hi -- using the latest DPP on Windows 8.1 I'm getting artifacts when rotating a small amount (0 to 2 degrees).
They are faint, and not normally visible, but they become important when doing techniques such as image averaging across many exported images, where you are often drawing lots of image detail out of the low level noise.
I created an animation of the effect at this link: http://caseyconnor.org/pub/image/rotation-example.gif
...that's a 26MB file so you will need to give it some time to load before it will play through smoothly. To create it, I exported multiple times from the same RAW image, altering the rotation by 0.1 degrees between -1.5 and 0 degrees. I then took the individual images and applied an extreme contast curve to make it more visible.
You can see a visible "grid" of dark lines that scales according to how much the image is rotated.
Past about 2 degrees, the grid isn't noticeable (because it scales so small as to be irrelevant), but of course 0-2 degrees is probably the most common range of rotation that will be used. I tried to come up with a reproducible method, and the best I have is:
- use an under-exposed image with a fair amount of noise (this pattern appears down close to the noise)
- use DPP's DLO feature at 100% with all other corrections disabled (CA, color blur, noise reduction, etc); the pattern appears without DLO enabled, but it's less obvious
- rotate a small amount (0-1.5 degrees)
- open in GIMP/PS/etc and apply a curve that boosts the shadows and draws out contrast in the low end
Again, it's not always obvious unless you are image-averaging (or image-median'ing) and pulling up the shadows a lot, in which case it becomes even more obvious, but it's serious enough of an issue that it has gotten in my way.
You can see an example of how bad it gets after image averaging (in this case a median calculation) here: http://caseyconnor.org/pub/image/bandingexample.jpg
The workaround is not to rotate in DPP, do the image averaging, and rotate in GIMP/PS/etc. But it would be nice if DPP could rotate properly.
Maybe it's a floating-point math rounding/precision error?
