RGB vs LRGB vs One-Shot-Colour

[MarkS]

Here are the results of a simulation I've been meaning to do for ages.  The idea behind the experiment is to compare results from a mono sensor to its one-shot-colour equivalent and, for the mono sensor, to compare RGB with LRGB acqusition.  I'm also interested in how much better LRGB might perform, compared to its OSC equivalent. This was done without using cameras but by using a test image, some maths and some numerical simulation.

I assumed 1 hour of imaging was done in each of 4 dithered positions. Dithering is a typical acquition technique and without using it the LRGB and OSC methods end up being significantly disadvantaged in this experiment due to spatial sampling problems.

So the assumption is:
1) For OSC we do 1 hour total exposure
2) For RGB we do 20min per channel
3) For LRGB we do 30min luminance followed by 10min each for R,G,B with 2x2 hardware binning.

A noise model was assumed where signal to noise ratio improves by square root of 2 for each doubling of exposure time - again this is typical of an imaging situation.

Here is the test object I used (one of my own images):


I added some artifical stars, as seen in this magnified crop:


These "stars" were added to make spatial sampling problems more apparent.

So straight to the results!
I recommend you right click each image and save them to your local machine so you can compare them, preferably at a scale of 200%

Here is the RGB result:


The LRGB result:


The OSC result:


From the perspective of apparent noise (and measured noise!) the RGB technique gives a noisier final result than LRGB or OSC.


Here are the magnified crops of the artificial stars:

RGB:


LRGB:


OSC:


The RGB result has best stars, as would be expected, since there are no spatial sampling issues. Their colour is perfect.
The LRGB result gives excellent results for black and white stars but some blurring of coloured stars. The smaller coloured stars also show some lack of colour.
The OSC result gives some blurring of all stars and the colour is very washed out for smaller stars.


I'm sure there are many other conclusions that could be drawn but for now, I'll present the results "as is".

Enjoy!

Mark

[Rocket Pooch]

WTF?

[Carole]

I would be really interested to see this test done using actual cameras and filters, also comparing it to a DSLR image given the same treatment. 

Carole

[MarkS]

WTF?

Cloudy nights!  Time on my hands.  I could always do some re-processing of a few of my images if you prefer!

[RobertM]

I would be really interested to see this test done using actual cameras and filters, also comparing it to a DSLR image given the same treatment. 

Carole

http://www.stargazer-observatory.com/2012/10/26/thoughts-about-monochromatic-and-osc-imaging-as-well-as-image-scale/

It's a very difficult thing to do unless you have both versions if the same chip.

Robert

[Rocket Pooch]

Mark,

It's just making me smile that's all, there's nothing in the posting that means anything.  It also looks like you have chucked in an assumption of focus point through filters for fun.  Unless the out of focus stars have something else going on.

I come form a slight scientific background there's no method or observation here just conclusion?  It's just a theory not based on fact.

But as you say, if you have nothing to do.

Chris

[MarkS]

I come form a slight scientific background there's no method or observation here just conclusion?  It's just a theory not based on fact.

Agreed, the explanation of the methodology was brief.  

The idea behind it was that for a given image captured using R,G,B filters with a mono camera (immaterial of how well the stars are focused) then it is deterministic and calculable the effects of directly substituting an alternative sensor configuration i.e. the use of a luminance filter or of binning or the effects of applying a Colour Filter Array to the sensor.  Clearly a CFA (Bayer matrix) is going to degrade the resolution of the image because of interpolation issues, the same applies for 2x2 binning of the chroma data.  The idea was to examine the precise nature of these degradations in a given test image.

Clearly some caution need to be used before applying the results of this to real-world imaging.  One obvious potential difference could be that the passbands of the RGB filters in the CFA of the OSC do not match those of the external RGB filter set being used with a mono camera.

Mark