With undersampling - will I get better image resolution with a mono compared to an OSC?

[festoon]

Currently I'm imaging at 135mm focal length with a an OSC sensor with a pixel size of 3.75 microns. Thus I'm undersampling at 5.73 arcsex/pixel. But I was wondering is my sampling even worse than this because I'm using an OSC with a Bayer Matrix? If so would I see a noticeable improvement in image resolution if I was to use a mono sensor with the same pixel size?

Edited February 25, 2021 by festoon

[wimvb]

You might see some slight improvement, but most likely not much. The deBayer interpolation algorithms do a good job in retaining as much detail as possible. Unless you only image planets and galaxies, of course. But for those you’d need a smaller pixel scale and much longer focal length anyway.

[festoon]

10 minutes ago, wimvb said:

Unless you only image planets and galaxies, of course. But for those you’d need a smaller pixel scale and much longer focal length anyway.

Deep sky imaging in general - I did take some images with this set up of M81 and M82 and as expected the resolution was poor. But I was thinking maybe some of that was down to the fact its an OSC sensor. Also from an inquistive point of view is there a way to calculate what sized OSC pixel size would give equivelent results to a mono setup?

[vlaiv]

1 hour ago, festoon said:

Currently I'm imaging at 135mm focal length with a an OSC sensor with a pixel size of 3.75 microns. Thus I'm undersampling at 5.73 arcsex/pixel.

Thing is - you are not under sampling.

I'm guessing you are using Samyang 135mm, right?

This is published chart by Samyang - even at F/8 - 30 lines/mm gives something like 95% contrast. 30 lpmm corresponds to pixel size of 33µm - you are using 3.75µm pixel size - almost 10 smaller.

When I was working with Samyang 85mm lens - also very sharp - it too can't handle pixel sizes that small. I calculated that good pixel size for such lens is about 10-12µm.

In another words - you'll see no difference between mono and color data. Only difference mono can make is using narrow band filters. Even then, lens is not diffraction limited but it is much closer to that performance.

[festoon]

Thanks @vlaiv. So I guess what you are saying is the lens is diffraction limited so having smaller pixels makes no difference. The thing is I saw a massive difference in perceived resolution moving from 6.5um pixels OSC to 3.75um pixels OSC. Obviously less blocky even on widefield targets.

Yes this is the Samyang 135mm f/2

Edited February 25, 2021 by festoon

[vlaiv]

10 minutes ago, festoon said:

Thanks @vlaiv. So I guess what you are saying is the lens is diffraction limited so having smaller pixels makes no difference. The thing is I saw a massive difference in perceived resolution moving from 6.5um pixels OSC to 3.75um pixels OSC. Obviously less blocky even on widefield targets.

Yes this is the Samyang 135mm f/2

I'm saying that lens is not even diffraction limited - in fact far from it.

Samyang 135 F/2 has aperture of 67.5mm.

In order to exploit 67.5mm F/2 diffraction limited optics - you need 0.5µm pixel size.

I would really like to see example where you see blockiness in image with 6.5µm pixel size and comparison to 3.75µm where you feel that sampling rate made big difference.

Here is what green channel of 4.8µm pixel size on Samyang 85mm F/1.4 lens stopped at F/2 looks like:

To my eye - that image starts to be properly sharp at about 1/2 - 1/3 of that resolution:

which equates to 9.6 - 14.4µm pixel size.

Mind you 85mm is not much worse than 135mm in terms of sharpness:

at least with sensor size I was using - diagonal less than 9mm, so distance from center is less than 4.5mm

[vlaiv]

Ok, yep, this is good comparison of what diffraction limited optics produces at these sampling rates:

80mm diffraction limited optics at 4"/px - same target:

We could argue that 4"/px is under sampling for that image - but I still don't see any blockiness in there.

[festoon]

I'll have to see if I saved any images with the 6.5um OSC. I certainly was not impressed hence why I switched to the ASI224MC Cool sensor with smaller pixels.

I guess undersampling may not be the correct word, but with the larger pixel size the stars appeared more square on the monitor.

Also purely from a understanding point of view I would like to understand if OSC does effect the resolution of the image

[vlaiv]

1 minute ago, festoon said:

I guess undersampling may not be the correct word, but with the larger pixel size the stars appeared more square on the monitor.

I'm not sure how that happens unless you use nearest neighbor resampling method and then zoom in (solution - use other resampling methods and don't zoom in )

2 minutes ago, festoon said:

Also purely from a understanding point of view I would like to understand if OSC does effect the resolution of the image

Depends how you process your data.

In principle - OSC camera really samples at half the sampling rate of mono camera and if you use super pixel mode for example to debayer your data - you'll get image as using pixels twice as large (sampling rate halved). If you use bilinear interpolation when debayering - you are not doing anything clever and it is the same as trying to get resolution by enlarging the image (again using bilinear interpolation) - you'll get larger image but not detail.

Only time when OSC camera can approach Mono in resolution for the same pixel size is if you utilize bayer drizzle for stacking. Unlike regular drizzle where one "assumes" pixel is smaller than it is (or better word is "pretends") - here we actually have "smaller" pixels - smaller in comparison to sampling rate. For that reason bayer drizzle is capable of providing almost the same resolution as mono sensor. You data needs to be properly dithered to do that.

In any case - difference between x2 sampling rate is not that big as you might expect, especially when you are close to optimum sampling rate.

Here is example of twice lower sampling resolution vs original.

I took original image that is already under sampled and then I reduced it x2. I then enlarged back that smaller image and created blink animation with original image.

You can see in smaller stars that some of sharpness is lost but in reality - not much of the detail is lost - at least not as much as you would expect (twice lower sampling rate sort of leads us to believe that there will be only "half of detail" - but that is not true).

With over sampled images - this difference is even less noticeable (in fact with over sampled images - you can get proper sampling by halving sampling rate - but even if you move from over sampled to under sampled - difference will be very small).

[festoon]

super info - thanks @vlaiv

So what I take from this is - when using an OSC if doing bilinear interpolation debayering sampling is x2 what it would be with a mono camera

However at x2 sampling some sharpness in smaller stars is lost, but image still shows good detail as shown in the GIF comparison you posted. I think the former is what I was experiencing with smaller stars

[andrew s]

There can be advantages with over sampling with CMOS cameras as it allows you to reduce telegraph noise . See here section 5.

Regards Andrew 

Edited February 25, 2021 by andrew s

[festoon]

Just a further thought on this...when I was using the 6.5um pixels at 135mm samyang then if sampling rate is x2 due to it being an osc, the the effective sampling is using an equivelent of 13um.

@vlaiv - do you not think that this will be starting to become noticeable in terms of undersampling as you calculated that good pixel size for such lens is about 10-12µm?

[vlaiv]

8 hours ago, festoon said:

Just a further thought on this...when I was using the 6.5um pixels at 135mm samyang then if sampling rate is x2 due to it being an osc, the the effective sampling is using an equivelent of 13um.

@vlaiv - do you not think that this will be starting to become noticeable in terms of undersampling as you calculated that good pixel size for such lens is about 10-12µm?

What do you think will be noticeable effect of under sampling?

I think that you'll be just fine with 6.5µm pixel size and Samyang lens if you do super pixel mode.

[festoon]

11 hours ago, vlaiv said:

What do you think will be noticeable effect of under sampling

The effect I am thinking is that too few pixels are captured for the resolution of the lens

[vlaiv]

42 minutes ago, festoon said:

The effect I am thinking is that too few pixels are captured for the resolution of the lens

I understand that - but how can you tell?

Do you think that this image is under sampled and why?

It is, by the way, sampling rate is 5"/px and it was taken with 8" telescope, but I'm wondering what in the image would make that obvious?

[festoon]

1 hour ago, vlaiv said:

Do you think that this image is under sampled and why?

At this zoom you dont see that it is undersampled. I understand that perception and presentation make a big difference. If you look at the stars and galaxy at a higher zoom it is apparent it is underampled

[vlaiv]

10 minutes ago, festoon said:

At this zoom you dont see that it is undersampled. I understand that perception and presentation make a big difference. If you look at the stars and galaxy at a higher zoom it is apparent it is underampled

If you zoom past 100% - all images are going to show artifacts. Point is - don't zoom beyond 100% unless you want to see artifacts. It has nothing to do with under sampling.

Look at these:

Top is nearest neighbor resampling, bottom is much more sophisticated resampling. Blockiness is not feature of under sampled image - it is feature of nearest neighbor resampling when you enlarge image. But you don' need to enlarge image using that algorithm - you can use B-splines or Mitchel or similar resampling method and get very nice looking smooth image (without much detail - after all you did zoom in 300%).

 

[festoon]

Point taken about zoom and how you sample or re-sample an image - dont go beyond 100% And I guess the point is it depends how you process your data i.e. if you re-sample use a method which smooths.

In terms of the specific lens I have (the Samyang 135mm) the measured resolution at the centre of the image plane at f/2.8 is 47 lpmm (as defined by MTF50, reference https://www.lenstip.com/442.4-Lens_review-Samyang_135_mm_f_2.0_ED_UMC_Image_resolution.html). That corresponds to a size of 21 um. To sample 47 lpmm sufficiently you would need a pixel size of 10.5 um or less. If you use a colour sensor (sampling reduced by x2) and you would require a pixel size of 5um or less not to lose detail.

Is this interpretation correct?

You also might argue that a vanishing resolution of an image could be defined at somewhere between MTF10 and MTF20. That would correspond to smaller and smaller sizes in terms of pixel size....at which point you come down to the seeing limit of perhaps 2 arcsec/pixel and a pixel size of 1.3 um.

 

[vlaiv]

11 minutes ago, festoon said:

Is this interpretation correct?

Yes, with addition that you should use super pixel mode for debayering OSC data so that you end up with 10.5 effective pixel size.

12 minutes ago, festoon said:

You also might argue that a vanishing resolution of an image could be defined at somewhere between MTF10 and MTF20.

In ideal lens, MTF10 would be somewhere at about twice the resolution of MTF50, however, camera lens are not ideal and don't have close to strait MTF curve.

Also at MTF10 - image will look soft and you would need to sharpen it up a bit - but that depends on effective SNR. Going with higher sampling rate lowers SNR that you can achieve in set amount of imaging time. There fore it's a balancing act - closer you are to critical sampling rate - more you can sharpen things up - but less SNR you have to sharpen it.

If you don't plan to sharpen things up - stay at about MTF50