Return of Interesting article on DxO about 5D III and D800...

Started Apr 10, 2013 | Discussions thread
schmegg MOD
Veteran MemberPosts: 5,257
Re: Return of Interesting article on DxO about 5D III and D800...
In reply to John Sheehy, Apr 14, 2013

John Sheehy wrote:

schmegg wrote:

However, acutance is not fixed at capture time like resolution is. It can be, and usually is, modified in post processing. And this is not considered at all in the MTF data because it is concerned with showing the relative sharpness of different lenses on a given camera - and changing acuity in post processing is of no relevance for the purpose of comparing lens sharpness. So not only do the graphs not tell the whole story, but they also show the opposite of what he claimed they show.

Hey, if RAW converters didn't heavily sharpen at the pixel level even when set to zero sharpening, everyone who pixel peeps for acuity would be very unhappy, even with an original 5D, or D30 or 1D.


I guess that DxO use their own converter for their analysis. But there's no guarantee that they don't apply different 'amounts' of conversion sharpening for the different cameras to start with.

That's why I personally prefer to keep comparisons from their data constrained to sensible bounds. Use their lens sharpness data to compare different lenses on a given camera for instance, but not use it to draw concrete conclusions about how different sensors will resolve detail in a final image.

It just isn't there, unless you omit the AA filter and have a very sharp lens, and then you get the gift of serious aliasing.  All RAWs from cameras with AA filters have low acuity at 100% pixel view.  IIRC, the typical range of maximum MTF at the nyquist for AA filters is 15-25, and for real lenses it is often well below that.  There is no problem raising the MTF of the conversion at the nyquist, except for noise.  A 32bit image could have its contrast decimated to 0.01%, and brought back to normal with no change at all in visible detail; you can do this in a program like ImageJ where you can blur a 32-bit image with a rectangular kernel the size of the entire image so that all you see is a solid gray, and then using the same rectangle to unblur it, you get an image indistinguishable from the original.  Throw a little too much noise in, however, before you reconstruct the original, and that little bit of noise becomes a noise-fest.

Cool stuff that.

Re noise - my assumption is that the sharpness tests are run at ISO100 and that no NR is used for them (I'd hope so anyway). So, in that respect, any noise present is included in the results.

Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow