[pics] SD10 vs. 20d sample

Started Feb 21, 2005 | Discussions thread
Helge Hafting Contributing Member • Posts: 827
Re: I sharpened your jpegs........

Peter G wrote:

PaulyOly wrote:

processing settings. Something I find strange is the difference in
the white sheet, the strands or texture looks larger in the sigma
shot, is this from upsampling?

No it is from Aliasing. Lack of an AA filter. It is the same
phenomena that produces 5 lines from 9 on the res chart in the
review.

Correct, but I wouldn't call it "lacking" for not having an AA filter. Bayer cameras don't have AA filters to blur out details, they have the AA filter for avoiding unsightly color aliasing. You will always get something wrong when taking pictures of details too small for the sensor. The "smudge it out" approach of an AA filter is no more "correct" than the aliasing we get without.

Either way, we get an image without the details the camera couldn't capture, and then the interesting question is "what looks best". For the cloth, the aliasing still looked like cloth and was therefore a good thing. The AA-filtered image simply looks unsharp. Of course one can find cases where blur look better than aliasing too, but those are rare. Most users of AA cameras fight the filter with unsharp masking.

As the details get smaller the actual ouput gets bigger.
This gives the impression that it is show more detail when in fact
it is obscuring the real detail and giving a bigger false pattern.

The false pattern is indeed false, but it does rarely obscure a real pattern you cannot capture anyway. The false blurring of an AA filter is often the bigger problem.

Sometimes people with a background in audio shows off formulas that "prove" why an AA filter of a certain strength is a good thing - or even necessary. But they work from a wrong assumption, the assumption that any false signal is very bad. This assumption holds for audio, where any false detail really kills the music. But the assumption rarely holds for vision, except in a few pathological cases where a chessboard pattern positioned "just so" turns into a set of broad stripes. Patterns a bit less regular (such as gravel, dirt and foliage) tends to alias into slightly coarser but similar patterns that are much more pleasing to the eye than the "mathematically correct" area of almost uniform color. Areas of uniform color is particularly bad when sharpening is applied later - you get a sharp line between the sky and the distant forest, but at such sharpness the eye really expect some approximation of distant trees and branches, instead of solid green.

As for the resolution chart, the sigma is clearly not showing a "true" image when it merges nine lines into seven and then into five. But at least it shows a set of black and white lines, where AA-equipped cameras shows a single column of gray. The sigma photo carries more information - "there is lines" where the AA photo merely says "don't know - too high resolution for me". There was no gray in the original resolution chart, so the gray column is "wrong".

Helge Hafting

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