In praise of AA filters

Started 10 months ago | Discussions thread
crames
Regular MemberPosts: 192
Like?
Re: One last thing ...
In reply to rovingtim, 10 months ago

rovingtim wrote:

Look, I'm not saying that Sigma Foveons never alias. There are other things happening to create artifacts that are strictly not aliasing. Without an AA filter, not only can detail from above Nyquist be aliased down, but also detail below Nyquist tends to be greatly boosted. Boosted detail in the range between Nyquist and 1/2 Nyquist often don't "look right" because of the sampling process itself, even though they are perfectly valid (not aliased). In order for those higher details to look right, it's necessary to invoke the part of the sampling theorem that is usually ignored with images: reconstruction. The zooming examples I gave are a way to "reconstruct" those higher details below Nyquiest.

I'm not a Foveon advocate, but I think they get a bad rap because anything that doesn't look right is blamed on the lack of AA filter.

Here is what someone points out about the AA:

http://www.dpreview.com/forums/post/52360078

The AA filter lowers pixel-to-pixel contrast in a very defined way. It does not "remove" detail, it lowers the detail contrast. Since it does this in a very well known fashion, it's rather easy to sharpen REAL detail back - in the camera without an AA filter the detail created by the de-Bayer algorithms have a much lower chance of actually being something "real" - as DSPo showed with the audio comparison.

This can be made visible with images too - look at the following image that contains first a de-bayered "raw-converted" full size 1:1 image, and then the "Red" and "Blue" channels directly from raw that created that image. They're also in 1:1 view - but since they're undersampled (only 1 in 4 pixels are either R or B) they're half as big as the de-Bayered image. They're the data that the raw-converter is supposed to build a "real" image from.

Except no raw converter nowadays works on the individual channels. They make use of inter- and intra-channel correlations.

Funny thing is, the AA filter works at twice the frequency it would need to to eliminate aliasing in the individual red and blue channels. AA filters are tuned to the luminance Nyquist rather than the "chrominance" Nyquist.

View: original size

View: original size

It's a small miracle that the raw converters of today can actually make something useful out of that, considering the huge amounts of moire visible in the raw file.

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