Sampling geometry and grain aliasing

[alanrockwood]

I have been trying to better understand grain aliasing, and I have a question.

Does anyone know what the sensor geometry is in a scanner? for example, does it look like the geometry indicated in the attached figure 1, which is sort of like a point-shaped sampling function or is it more like the geometry in the attached figure 2, which is a fat-shaped sampling function?

If the sampling function resembles the first figure then the noise in the reconstructed image will be enhanced (grain aliasing). If the sampling function resembles the second figure then there will be less noise enhancement (less grain aliasing.)

I will probably follow up with a comment explaining these conclusions, but for now please feel free to comment on what I have written so far. Comments on the shape of the sensor array are especially welcome.

 

[koraks]

I expect option 3: overlapping peaks.

 

[Adrian Bacon]

Grain aliasing only happens in situations where the spatial nyquist sampling frequency of the sensor is less than the actual grain size AND there is no optical (or otherwise, can also be diffracted, or otherwise just not that optically sharp) filter that ensures any spatial frequencies that are higher than the sampling frequency don't make it through to the sensor.

So, in your diagrams, both sampling functions would generate grain aliasing. The first one would be horrendous. The second one less so, but still pretty nasty. A good scanning system optically would generate what @koraks said: overlapping peaks. One grain would be blurred (or diffracted, or just not that sharply focused) that it would never be smaller than one sampling pixel, and no matter where it was, it would always fall on either one sampling pixel directly, or would span two or more sampling pixels.

A lot of relative noise is made about grain aliasing, but the reality of the matter is, it only really shows up in very badly designed optical systems, or the software is doing something after the sampling stage like taking every 2nd, 3rd, (or more) pixel to downscale instead of doing a proper resampling from the higher pixel density to lower resolution.

Much of what people say is grain aliasing really is just badly handled scanning. What you do with the samples after they've been acquired matters just as much as the actual acquisition of said samples.

 

[Eric Rose]

Are you seeing any problems in your prints?