Monochrome r/g/b images from raw?

[dreed]

It is still possible to get images with aliasing even if you don't perform interpolation, so the M Monochrom will not be inmune to this.
Aliasing happens when the frequency of the signal is above Nyquist frequency of the sampling device.

From today's review by Michael:


But, there's more going on. To create a colour image a Bayer sensor is, in fact, "guessing"; or interpolating between the various cells to recreate a colour rendition. As a consequence, down at the pixel level there is a "fudging" going on. This is what leads to the possibility of colour moire, when an interference pattern occurs between the Bayer array and very fine patterns in the real world (fabrics, bird feathers, etc). Even when you later take away colour in post processing there are still interference artifacts remaining.

The Leica Monochrom M has none of these issues, because there is no Bayer array. Therefore actual resolution is higher that a Bayer sensor with the same number of pixels, and there is no possibility of either colour moire or related artifacting.

This I find quite interesting. I suppose I'm just going to need to find a subject that produces moire at 100% and test this out for myself.

[FranciscoDisilvestro]

Well, being Monochrome it will not produce "colour" moire for sure, but it will definetly be possible to get aliasing, it is basic digital signal processing theory.

To illustrate this, the first chart shows a signal to be sampled together with a sampling signal with too low frequency. The second chart shows the samples obtained and a recostructed ouput signal, which is at a lower frequency compared to the origial. This is basically what happens, the output does not resemble the original signal (aliased).

[dreed]

Well, being Monochrome it will not produce "colour" moire for sure, but it will definetly be possible to get aliasing, it is basic digital signal processing theory.

Look, I don't even need a camera to create moire - I can easily write a Java program to draw a problematic pattern on the screen very easily. All of those test images will produce moire, in one fashion or another, on a computer screen if they're drawn without aliasing techniques. A camera isn't required to produce moire.

What I'm interested in is what happens in real life, not on some special wheel pattern that is known to produce moire. That's uninteresting.

What I want to know is what (say) the raw red channel looks like from an image that has been taken of (say) a building with vertical stripes that when looked at with 100% magnification in colour, exhibits moire.

[BartvanderWolf]

What I'm interested in is what happens in real life, not on some special wheel pattern that is known to produce moire. That's uninteresting.

Hi,

It's not the pattern that produces the moiré, it's the undersampling by the sensor that does, just as with real life subjects. The predictable pattern just makes it easier to see, because everything that deviates from the expected pattern will stick out like a sore thumb. Even random (non-repetitive) patterns will have aliasing, you just won't notice it, unless the color deviates from the expected.

What I want to know is what (say) the raw red channel looks like from an image that has been taken of (say) a building with vertical stripes that when looked at with 100% magnification in colour, exhibits moire.

In this thread Guillermo explains how you can use the non-demosaiced output from DCRaw together with a 50% nearest neighbor downsampling to isolate the single channels. You'll just have to figure out which channel it is you're looking at before/after adding a column/row before downsampling. Obviously the red channel will show red object colors as brighter (after whitebalancing) than green and blue.

Cheers,
Bart