IR lucky imaging of long-lens landscapes

[Øivind Tøien]

My recent technical foray into asto cam imaging with a small sensor and lucky imaging made me wonder what a similar technique as used on the moon could do for long-distance landscapes when combined with IR imaging. The three high mountain peaks of the Alaska range, Mont Hayes, Hess Mountain and Mont Deborah more than 150km south of Fairbanks seemed like a suitable target. The day was clear, but with quite a bit of haze almost hiding the mountains. Here is a cropped view from the webcam at the Geophysical Institute, simulating the view with a normal lens. The arrow points to Hess Mountain. Mont Hayes on the left side and Mont Deborah next to Hess Mountain at the right side.
#1



So I mounted all the lens power I have, which is the 300mm f/4 PF with TC-14E + TC-20E III stacked. First a full frame of Hess Mountain from D500, contrast turned all the way up in CNX-D, and with as much sharpening that I could apply without creating too much artifacts:
#2



A crop of this frame similar to that used for the asto cam reveals problems with optical quality of the air between the lens and target:
#3



The the same frame of Mont Hess captured with the ASI678MC astrocam at a 2um pixel pitch with a 685nm IR pass filter. Stack of the best 25% of 2000 video frames. The angle of view is about 0.6° which is about the same as a 4000mm lens on a full frame body.
#4



A view of the 4216m tall Mont Hayes with similar technique also demonstrates how the IR pass filter cuts through the haze. The lucky imaging can only do so much though. There are notable parts that are more blurry than others.
#5



Then finally a stich of two horizontal frames of Mont Deborah's peak to reveal a little more of this 3761 m tall mountain. The elevation at my vantage point in Fairbanks is only about 200m.
#6

[Birna Rørslett]

Oooh, nice. We'll all have to go get astro cams now

I'm positively surprised this much of optical quality remains after adding two TCs to the master lens.

[Øivind Tøien]

Yes, it is quite surprising- but then there is the stacking and wavelet sharpening that helps bring out the last details. The resulting f/ratio of 11 with the lens wide open is supposed to be well matched to the 2uM pixel pitch according to common advices in the planetary imaging fora, it is quite common to use 3x Barlows (tele extenders) on the scopes. But I discovered later that I had the lens stopped down 2/3 stop in this case (forgot to turn the D500 off after that first capture when dismounting the D500, so the the aperture used for that remained), so there would be more diffraction than optimal for the 2um pitch.  It is not a very practical setup for landscape imaging though, depending on a computer for the recording and each of those 2000 frame videos takes about 16GB!

[Akira]

So, the first image is already an equivalent FOV of 1280mm?  That is amazing!  This is a great way to utilize an astro camera combined with an IR filter.  Thank you for sharing.

[John Geerts]

Amazing results,  Øivind

[Øivind Tøien]

Thanks Akira and John. Yes the lens plus stacked TCs combo is effectively an 840mm f/11 telephoto lens, which on the D500 sensor in image #1 has the same angle of view as a 1260mm lens on a full frame sensor.

[Bernard Delley]

Yes, it is quite surprising- but then there is the stacking and wavelet sharpening that helps bring out the last details. The resulting f/ratio of 11 with the lens wide open is supposed to be well matched to the 2uM pixel pitch according to common advices in the planetary imaging fora,

But I discovered later that I had the lens stopped down 2/3 stop in this case (forgot to turn the D500 off after that first capture when dismounting the D500, so the the aperture used for that remained), so there would be more diffraction than optimal for the 2um pitch.  It is not a very practical setup for landscape imaging though, depending on a computer for the recording and each of those 2000 frame videos takes about 16GB!

quite amazing results!
The f/14 leads to an optical cutoff frequency of 104.3 c/mm for 685 nm light. So with a monochrome sensor with 2 mum pixel pitch you got quite a bit oversampling (~5x) on the corresponding diffraction limited image.
For the Bayer red of the D500 you remain under sampled at the diffraction limit. I would be curious anyway, on how the D500 would render with the  685 nm filter. A lucky imaging with the 4 best images out of 16 would approximately double the SNR.

[Øivind Tøien]

quite amazing results!
The f/14 leads to an optical cutoff frequency of 104.3 c/mm for 685 nm light. So with a monochrome sensor with 2 mum pixel pitch you got quite a bit oversampling (~5x) on the corresponding diffraction limited image.
For the Bayer red of the D500 you remain under sampled at the diffraction limit. I would be curious anyway, on how the D500 would render with the  685 nm filter. A lucky imaging with the 4 best images out of 16 would approximately double the SNR.

Thanks so much for your comment Bernard. You are touching an interesting question regarding how we consider the diffraction to affect resolution in general photography vs. what is the going advices given for planetary astrophotography imaging. Can you remind us how you arrive at 104.3 c/mm (assuming c means cycles, btw, my ASI678MC sensor is not monochrome but has a regular RGGB color matrix).

I have seen a number of threads on the Cloudy Nights planetary forum where the optimal f-ratio, usually recommended to be 5x pixel pitch for visible light imaging, is hotly debated. But it seems to be what is generally accepted and in the FAQs, although IR imaging should modify this number. However I have not seen the actual physics behind it too well explained. The threads typically will refer to the formulas on this page, https://www.cloudynights.com/topic/413414-planetary-imaging-rule-of-thumb-which-barlow/, but it is not explained where the constant used comes from. Generally it seems that much smaller f-ratios are considered optimal in planetary imaging than where we usually consider diffraction to become a problem in general photography at a given sensor pitch  My current intuitive understanding (which might not be correct) is that there are interplays between diffraction limiting, the advantage gained by stacking (i.e. ability to  amplify details with minute differences in contrast by wavelet sharpening without getting too noisy), the limitations imposed by seeing (optical quality of the atmosphere) and the desire to get as much magnification as possible. So if one has f/5 scope or lens, one would add Barlows/teleconverters to increase magnification until the f-ratio is f/10 with a 2 um color sensor pitch.   

[Bernard Delley]

Can you remind us how you arrive at 104.3 c/mm (assuming c means cycles, btw, my ASI678MC sensor is not monochrome but has a regular RGGB color matrix).

I have seen a number of threads on the Cloudy Nights planetary forum where the optimal f-ratio, usually recommended to be 5x pixel pitch for visible light imaging, is hotly debated. But it seems to be what is generally accepted and in the FAQs, although IR imaging should modify this number. However I have not seen the actual physics behind it too well explained.   

The formula for the optical cutoff frequency for a circular aperture is simple,  f_cutoff = 1 / ( N * lambda ) ;  here N=14  and lambda = 685 nm . The contrast transfer starts at 100% for frequency zero and goes down a bit faster than linear towards the cutoff frequency with a little tailing off towards the cutoff above. The contrast for higher frequency content is strictly zero beyond the cutoff.  The denominator in that calculation is ~ 10 micro meters.   One could wish that no aliasing can happen  in the pixel detector recording all that optical information. So the Nyqvist frequency for the pixel sampling should be equal or greater than the cutoff frequency from the aperture. For the horizontal or vertical direction of green (or monochrome) pixels one would thus like the pixel pitch a factor 2x smaller than the denominator.    If one considers the unfavorable 45 degrees direction, one would need a sqrt(2) x smaller pixel pitch : 3x smaller than denominator. The horizontal direction for red or blue Bayer pixels is unfavorable requiring 4x smaller than the denominator. Smaller pixels are better provided the read noise goes down with the pixel area. An easier condition may be that also the full well capacity should not go down too fast. So 5x is a compromise suggested by the practitioners: for lambda = 400 nm  and f_Nyqvist_monochrome_horiz = 2 you get
N = 5x pixel_pitch [ micrometer]

[Øivind Tøien]

Thanks for the explanation, Bernard. I guess the stacking allows to pull out details beyond the normal contrast cutoff with temporal oversampling, but of course a lot of that is to overcome the shortcomings due to limited seeing. The mountain seen above is severely limited by seeing. Object high in the sky will have a lot less atmosphere to potentially mess up resolution. If the wavelet used on the mountains was applied on a solar or lunar capture, it would be heavily oversharpened if seeing is reasonably good. Here is an optimal processing of a recent capture of active sunspot regions AR3288 and AR3285 to the right, with Baader Astrosolar safety film and UV-IR cut filter. The grainy structure is not noise - notice how its blurriness varies across the surface of the sun: