With all the discussions on AA filters what aliasing artifacts are there other than moire? can someone post a few clear examples? I was under the impression (misunderstanding?) that moire was caused because you could sample a repeating pattern with a sensor that has a repeating pattern at greater than the Nyquist frequency. But all the talk about the D800E got me to thinking in landscape photography pictures of trees, foliage etc. without a pattern would there be false information present?
Hi Marc,
Any detail that is smalller than 2 sensels will aliase when sampled with a regular grid without proper low-pass filtering. The artifacts are most pronounced when sampling regular repetitive structures, because that produces moiré and that is often a structure that we humans can see easily because the human visual system (HVS) tends to look for patterns (which reduces the visual information to manageable amounts to process in time). Anything that deviates from our expectations raises a red flag (fight or flight instinct gets alerted).
Sharp edges also have detail that's too small for our sensors to resolve
reliably(!). In digital signal processing (DSP) circles is is common to interpret sharp edges as a combination (an addition) of Sine or Cosine wave patterns of various frequencies and it requires also extremely fine (high frequency) waves to approximate those sharp edges. It would also require extremely dense sampling to be able and achieve that in an image.
If the sampling density is not high enough, then e.g. edge signal will snap to the nearest sensel, and not to the neighboring sensel, thus causing jaggies, stairstepped edges instead of smooth anti-aliased edges which are closer to the true representation of the scene. So if branches, or powerlines, or roofedges, or text, or other high contrast edges and lines are not resolvable, not sampled with a dense enough sampling grid, then blocking, stairstepping artifacts wil become visible, especially noticeable when enlarging to well within the human visual acuity (or Vernier acuity which is much higher than 20/20 vision). Sharpening will only make the artifacts stand out more.
To worsen the potential problem, our Bayer CFAs sample Green at a higher density than Red and Blue. Therefore Red and Blue will aliase more than Green does and that creates a problem for the Demosaicing, and can result in False Color artifacts. That are colorful pixels especially noticeable on high contrast edges (e.g. dark branches or treetrunks or powerlines against a light background of white clouds).
Software can only reduce the false color artifacts a bit (e.g. by desaturating edges), but it cannot decide what is real detail and what is stairstepping, without human intervention (blurring at the Raw level). And blurring will not solve all issues, because aliases are larger representations or the real detail, sometimes much larger. These larger representations of very fine detail cannot be separated from the larger real detail, there is not enough information left to do that reliably. The only solution to prevent aliasing is a low-pass filtering of the signal before it is sampled.
So, in the absence of proper low-pass filtering, stairstepping and blocking, and false color artifacts (most noticeable on high contast edges and lines), are the things to look out for when there are no repetitive stuctures to cause moiré.
Cheers,
Bart
