In praise of AA filters

Started 10 months ago | Discussions thread
John Sheehy
Forum ProPosts: 16,019
Like?
Re: Wait a minute: Foveon example not so aliased after all...
In reply to crames, 10 months ago

crames wrote:

John Sheehy wrote:

crames wrote:

Take a look at the foveon example zoomed 300% with sinc interpolation. There is actually only a few small aliased patches, mostly in the upper left.

Huh? I see moire in the right edge, second patch from bottom, the right bottom of the first full patch in the lower right corner, the patch just a little left and south of center, and I also see inconsistent rendering of the darker spots in the areas where the three-toned lines cross the other lines, due to luck of alignment. This is a clearly aliased capture. I don't see just fabric here; I see poor capture, which ruins the fabric. Face it, we need cameras with a hell of a lot more pixels to properly record things like this. Shortcuts do not work.

Yes, I said "mostly" in the upper left. It's nice that you were able to find some artifacts in other areas, but without knowing the origin and processing history of this image it's hard to say what is the exact cause of some defects. For sure I would say that the moire that occurs at angles that are inconsistent with the direction of the weave is aliasing.

All moire is aliasing.  All aliasing is not moire, though.  Moire is an extra pattern caused by the juxtaposition of existing patterns.

The point is that most of the "aliasing" goes away with a little interpolation, which means that most of the so-called aliasing is in fact not aliasing.

It doesn't go away.  You lower the contrast of everything

Lee Jay said that the image is "covered with aliasing." "The entire thing is full of jaggies, false detail and patterns that don't exist."

Here is the original X3 image again, this time zoomed with nearest neighbor.

Yes, and the aliasing, including the moire, is much easier too see, as original pixel transients are maintained at the edges of the big tiles.

There you go, jaggies, "snap to grid," moire patterns, etc.

Here it is with Photoshop bicubic smoother @300%:

Would you say that it is still covered with aliasing, or would you agree that most of the artifacts have disappeared? True aliases do not go away so easily.

I would say that the contrast of all small detail, real or artificial, is now much lower, but I can still see patches of higher contrast and lower contrast areas, due to luck of alignment between sensor and subject detail.  I also see softened jaggies, not the lumpless near-analog texture we would have had with 9x as many pixels.

Not in a 1:1 display; then there has to be a difference in capture for there to be a difference in output. A converter does not randomly increase micro-contrast in some areas and decrease it in others.

I'm not understanding what you are saying here. Reconstruction artifacts are especially visible in a 1:1 display, since no reconstruction occurs, other than representing pixels as little sharp-edged squares or rectangles, a distortion that occurs regardless of whether the image was captured properly or not.

I'm saying that if we are looking at a 100% view, then the low-frequency moire artifacts could not have come from processing.  You can only have modulating pixel contrast if the capture is aliased.  Perhaps you have a much narrower definition of aliasing than I do.  Do you believe that there is a range of sampling artifacts occurring between oversapmpling and your definition of aliasing, which are caused by undersampling but not "aliasing"?  IMO, every dot, and every edge in an image should have the same visible blur regardless of how the sensor aligns with the subject; anything less than this is "aliasing".  Proper sampling is oversampling, IMO.

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