AA filter question - Joe?

Started Feb 9, 2012 | Discussions thread
hjulenissen Senior Member • Posts: 2,139
Re: AA filter question - Joe?

There are some extreme cases that may be enlightening.

Imagine a (part of) a scene that mainly excite the "red" channel. With no AA-filter and a perfect lense/technique, even with 100% fill-rate, the red sensel will sample its channel in a 2x2 decimated pattern.

If the scene happens to contain a pattern that oscillate at exactly fs/2 (+1 at each red sensel, 0 at each non-red sensel), the CFA output will look like a constant, highly saturated motif. I dont see how any clever debayer/demoire algorithm can cure that without destroying other scenes that happen to contain flat, highly saturated parts.

This is the core of aliasing - two very different scenes may give the exactly same raw file. Once that raw file is recorded, how could a raw developer know which one of those potential scenes was the real scene? Of course, it can make educated guesses, and the Photoshop jockey may be able to help it make guesses, but that does not appeal to me.

My argument earlier was that the benefits of removing the OLPF should be smaller and smaller as the number of pixels increase - while the drawbacks will still be constant (assuming perfect lense, technique etc).

hjulenissen wrote:

...Also, with increasing pixels counts (and larger pass-band OLPF), the high-frequency behaviour will probably be less and less interesting.

But the effects of aliasing wraps around to lower frequencies. So in that future 1 gigapixel camera, who would want to risk highly visible artifacts in low spatial frequencies in order to have a little more gain at fs/2??


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