Debayering a DSLR's Bayer matrix.

[Adam J]

Ok I have a set of comparison images. Some baby food bottles on the side in my kitchen but it will have to do.

Each image is a 1/50th of a second exposure through my Baader 2" 7nm filter fitted to the front of a 50mm lens ISO800 F1.8 (although as discussed short forcal lengths may have an effect on the mono sensitivity).

From left to right: 1) Mono Camera Single Red Channel Only Extracted. 2) Mono Camera all four pixels extracted (2x2). 3) RGB Camera Red Channel Only Extracted (Green and Blue Removed).

I think that the clear winner in this test is the original OSC camera it simply has the best signal to noise ratio. Now I will still be doing the real world test on a Ha nebula as soon as possible as you could convince me that this may be due to the short focal lenght of the lens....but I would not bet on it.

You could also convince me that ultimately with longer exposures / integration I would get more detail out of the mono when processing it for full resolution as opposed to binning it. But from what I can see here I would expect to have to increase the integration time by about a factor of four to achieve the same signal to noise level as in the Red channel only of the RGB camera.....I have seen this done by someone with a mono camera and the results are very good but the integration times are crazy (10's of hours). This leaves me of the opinion that the camera is (baring a a big change in the real world test) not going to give me what I am looking for and I would be better off taking H-a and OIII data with the bayer intact.

Edited November 19, 2016 by Adam J

[Thalestris24]

Hi

It seems to me that it would only make sense to image a Ha source when using the Ha filter and then processing the image taken by a mono sensor in it's full resolution. Similarly with an Oiii source/filter.

Louise

[Adam J]

2 minutes ago, Thalestris24 said:

Hi

It seems to me that it would only make sense to image a Ha source when using the Ha filter and then processing the image taken by a mono sensor in it's full resolution. Similarly with an Oiii source/filter.

Louise

With the filter in place the same amount of 656.28nm (+ or - 3.5nm) light is getting through the H-a filter and reaching each of the sensors. With the H-a filter in place the bottle in the image is effectivly a h-a source in so far as the camera is concerned. A full resolution image will have the same signal to noise as image 1 but at a higher resolution.

[Xplode]

6 hours ago, GTom said:

It's all very nice, however 12" f/3.8 would probably yield yaw-dropping results with Bayer-on too. I see you have both "color" and mono 6D, do you have a comparable pair, where only the camera is the changing factor? For a 4-5min exposure each I guess the same scope would do, with camera swap.

Do you experience more vignetting on the mono version with photo lenses (f/2 for instance)?

I haven't tried it with really fast lenses so i can't say if it has had any affect on vignetting. 

Unfortunately i don't have anything to compare, it's all taken with different filters or scopes and i don't actually have both cameras anymore so i can't make any comparisons either

The 6Da was sold a few weeks ago and the 6Dm was just sold the other day.

Wanting to use a filter wheel on the newtonian i could not use the 6Dm so it had to go to allow me to buy a CCD with filter wheel.

What i can do at a later point is compare data from the CCD camera with the 6Dm data with the same filters and exposure time.

 

30 minutes ago, Adam J said:

Ok I have a set of comparison images. Some baby food bottles on the side in my kitchen but it will have to do.

Each image is a 1/50th of a second exposure through my Baader 2" 7nm filter fitted to the front of a 50mm lens ISO800 F1.8 (although as discussed short forcal lengths may have an effect on the mono sensitivity).

From left to right: 1) Mono Camera Single Red Channel Only Extracted. 2) Mono Camera all four pixels extracted (2x2). 3) RGB Camera Red Channel Only Extracted (Green and Blue Removed).

I think that the clear winner in this test is the original OSC camera it simply has the best signal to noise ratio. Now I will still be doing the real world test on a Ha nebula as soon as possible as you could convince me that this may be due to the short focal lenght of the lens....but I would not bet on it.

You could also convince me that ultimately with longer exposures / integration I would get more detail out of the mono when processing it for full resolution as opposed to binning it. But from what I can see here I would expect to have to increase the integration time by about a factor of four to achieve the same noise level as in the Red channel only of the RGB camera.....I have seen this done by someone with a mono camera and the results are very good but the integration times are crazy (10's of hours). This leaves me of the opinion that the camera is (baring a a big change in the real world test) not going to give me what I am looking for and I would be better off taking H-a and OIII data with the bayer intact.

Nice test Adam

It agrees with what i've been saying earlier that the biggest gain isn't in narrowband imaging, but when used for luminance.

Which F value was the comparison done at? You should do a test like with with different F values and with both the Ha og OIII filter.

Also making an image where the 2x2 is used as a superpixel would be great.

[Thalestris24]

5 minutes ago, Adam J said:

With the filter in place the same amount of 656.28nm (+ or - 3.5nm) light is getting through the H-a filter and reaching each of the sensors. With the H-a filter in place the bottle in the image is effectivly a h-a source in so far as the camera is concerned. A full resolution image will have the same signal to noise as image 1 but at a higher resolution.

That's mostly true but in practice the filter is only used to image emission nebula, and that obviously needs much longer exposures. With bright sources the filter won't block a lot of the non-Ha light.

[Adam J]

47 minutes ago, Thalestris24 said:

That's mostly true but in practice the filter is only used to image emission nebula, and that obviously needs much longer exposures. With bright sources the filter won't block a lot of the non-Ha light.

Your right it will let a small amount if broad band light past the filter.

However, I already considered this, due to the narrower pass band of the single red pixel in the RGB camera it will let allot less broad band light through to the detector than in the case of the mono which is seeing everything. So yes its not a 100% pure test, but it is a test that is in fact biased in favor of the mono camera not the RGB camera. As a result it did not concern me as its likely to make the mono appear better than it really is in comparison to the RGB not worse and hence would not change the result.

The second thing that makes me suspect that this is not an issue is that the signal level in the blue and green channels of the RGB image was a good 20dB below the peak of the red channel. See BackyardEOS histogram below, indicating that it is mostly noise.

In this experiment I am looking at signal to noise. In this case the noise is mostly going to be bias / read noise as the exposures are very short. However, while the signal to noise ration in any given image is likely to change with different exposures (at least until the thermal noise becomes dominant) the ratio of the signal to noise ratios (that is a mouthful) between the two cameras is likely to remain fixed.

I am 100% open to being wrong on this still, or I would not be planning to do the real world test against an emission nebula. I am just stating the facts based on testing up to this point. I have not finished my analysis yet.

 

48 minutes ago, Xplode said:

I haven't tried it with really fast lenses so i can't say if it has had any affect on vignetting. 

Unfortunately i don't have anything to compare, it's all taken with different filters or scopes and i don't actually have both cameras anymore so i can't make any comparisons either

The 6Da was sold a few weeks ago and the 6Dm was just sold the other day.

Wanting to use a filter wheel on the newtonian i could not use the 6Dm so it had to go to allow me to buy a CCD with filter wheel.

What i can do at a later point is compare data from the CCD camera with the 6Dm data with the same filters and exposure time.

 

Nice test Adam

It agrees with what i've been saying earlier that the biggest gain isn't in narrowband imaging, but when used for luminance.

Which F value was the comparison done at? You should do a test like with with different F values and with both the Ha og OIII filter.

Also making an image where the 2x2 is used as a superpixel would be great.

Yep I am seeing better performance from the Mono when the CLS filter is used. Would be good for higher resolution images of galaxies. But I am starting to feel its going to be better to use the RGB camera for narrow band.

I think that there is a new coma corrector on the marked that increases back focus enough that it allows you to use a filter wheel with a DSRL on a fast newt without effecting the focal length. https://www.firstlightoptics.com/coma-correctors/baader-rcci-rowe-coma-corrector.html

Edited November 19, 2016 by Adam J

[diegocolonnello]

I did it!!!!!!!!!!! so happy

 

Canon 40Da-M

hahahahaha

 

[GTom]

Nice test Adam, very interesting: clearly more intensity when all pixels are used but less details. Latter goes in the opposite direction as Maxmax's marketing, claiming that monochrome should yield more details.

Can you share the raw files?

I don't really see  loss of intensity on the subpixel level (#2 vs #3), maybe 1/3stop, whereas the gain on the additional pixels is substantially more, my guess is around 1EV.

 

12 hours ago, Adam J said:

Ok I have a set of comparison images. Some baby food bottles on the side in my kitchen but it will have to do.

Each image is a 1/50th of a second exposure through my Baader 2" 7nm filter fitted to the front of a 50mm lens ISO800 F1.8 (although as discussed short forcal lengths may have an effect on the mono sensitivity).

From left to right: 1) Mono Camera Single Red Channel Only Extracted. 2) Mono Camera all four pixels extracted (2x2). 3) RGB Camera Red Channel Only Extracted (Green and Blue Removed).

I think that the clear winner in this test is the original OSC camera it simply has the best signal to noise ratio. Now I will still be doing the real world test on a Ha nebula as soon as possible as you could convince me that this may be due to the short focal lenght of the lens....but I would not bet on it.

You could also convince me that ultimately with longer exposures / integration I would get more detail out of the mono when processing it for full resolution as opposed to binning it. But from what I can see here I would expect to have to increase the integration time by about a factor of four to achieve the same signal to noise level as in the Red channel only of the RGB camera.....I have seen this done by someone with a mono camera and the results are very good but the integration times are crazy (10's of hours). This leaves me of the opinion that the camera is (baring a a big change in the real world test) not going to give me what I am looking for and I would be better off taking H-a and OIII data with the bayer intact.

Edited November 19, 2016 by GTom

[Adam J]

2 hours ago, GTom said:

Nice test Adam, very interesting: clearly more intensity when all pixels are used but less details. Latter goes in the opposite direction as Maxmax's marketing, claiming that monochrome should yield more details.

Can you share the raw files?

I don't really see  loss of intensity on the subpixel level (#2 vs #3), maybe 1/3stop, whereas the gain on the additional pixels is substantially more, my guess is around 1EV.

 

Ah not its not a test for intensity these have been processed to attempt to examine the signal to noise ratio, i.e how much detail can be extracted from the image.....although as the noise level is almost identical between the two cameras you could in fact. say that the level of detail is a indication of intensity or to be more exact the signal level itself.

 

I have never shared RAW files before as they are quite large. Ill take a look into how I might do that. I am assuming drop box or something like that, although I dont actually have an account.

 

I am going to try super pixel mode on the mono data next.

Edited November 19, 2016 by Adam J

[GTom]

Right, I thought Mediafire doesn't need a registration for uploading, but apparently it does.

[Adam J]

I am probably not going to post on this again until I can get some real world comparison images. However you put it that is the only test that is going to actually matter. Am open to suggestions about targets and exposure settings though.

Am also of two minds as to if I should cool the cameras in the cool box for the test. Problem is that as I only have the one cool box it will mean a long delay in changing between cameras. So I am thinking ill just do the test at ambient temperature.

Edited November 19, 2016 by Adam J

[Herra Kuulapaa]

15 hours ago, Adam J said:

Ok I have a set of comparison images. Some baby food bottles on the side in my kitchen but it will have to do.

Each image is a 1/50th of a second exposure through my Baader 2" 7nm filter fitted to the front of a 50mm lens ISO800 F1.8 (although as discussed short forcal lengths may have an effect on the mono sensitivity).

From left to right: 1) Mono Camera Single Red Channel Only Extracted. 2) Mono Camera all four pixels extracted (2x2). 3) RGB Camera Red Channel Only Extracted (Green and Blue Removed).

I think that the clear winner in this test is the original OSC camera it simply has the best signal to noise ratio. Now I will still be doing the real world test on a Ha nebula as soon as possible as you could convince me that this may be due to the short focal lenght of the lens....but I would not bet on it.

You could also convince me that ultimately with longer exposures / integration I would get more detail out of the mono when processing it for full resolution as opposed to binning it. But from what I can see here I would expect to have to increase the integration time by about a factor of four to achieve the same signal to noise level as in the Red channel only of the RGB camera.....I have seen this done by someone with a mono camera and the results are very good but the integration times are crazy (10's of hours). This leaves me of the opinion that the camera is (baring a a big change in the real world test) not going to give me what I am looking for and I would be better off taking H-a and OIII data with the bayer intact.

Hmmm., there is something funny in your test. It might be the sensor too, because Exmors seem to behave very nice after mono conversion. They have much more narrower conductors than 1000D though, which may decrease the amount of photons hitting them.

Here is some real world testing

Here is a comparison of full spectrum RGB D5100 on left vs D5100am (astromono) on right. Both files are true raw (non debayered).

click it large

Edited November 19, 2016 by Herra Kuulapaa

[Stub Mandrel]

22 minutes ago, Herra Kuulapaa said:

click it large

Interesting. Most of the difference would appear to come from the debayering process failing to completely remove the pattern noise of the RGB sensor. Presumably this could be tuned out?

 

[GTom]

Nice clear + apparently a bit more details than the RGB. Do you still have both cameras? Can you do a comparison with narrowband filters (H-a + O-III) and a fast, maybe f/2 lens?

I am awaiting a commercial offer for the conversion of my a6000, but I fear that the final costs (a6000 resale value, conversion, cold-finger cooling+ all shipping&taxes) will go beyond the price of an ASI1600mm - the next question would then be if the tortured a6000 is any better, which I doubt...

 

26 minutes ago, Herra Kuulapaa said:

Hmmm., there is something funny in your test. It might be the sensor too, because Exmors seem to behave very nice after mono conversion. They have much more narrower conductors than 1000D, which may decrease the amount of photons hitting them.

Here is some real world testing

Here is a comparison of full spectrum RGB D5100 on left vs D5100am (astromono) on right. Both files are true raw (non debayered).

click it large

Edited November 19, 2016 by GTom

[Herra Kuulapaa]

11 minutes ago, GTom said:

Nice clear + apparently a bit more details than the RGB. Do you still have both cameras? Can you do a comparison with narrowband filters (H-a + O-III) and a fast, maybe f/2 lens?

Unfortunately I've sold the mono and converted RGB to cooled so it's not possible to conduct more tests at this point.

I must say that the demosaicing algorithm is very good to replicate almost same detail level. The true difference comes in narrow band where all the four matrix pixels will record signal same way.

Here is a small comparison of the same full spectrum target in true raw:

RGB:

Mono:

Edited November 19, 2016 by Herra Kuulapaa

[GTom]

That is quite a difference in intensity! Were they identically exposed? (time, aperture, target)

1 hour ago, Herra Kuulapaa said:

Unfortunately I've sold the mono and converted RGB to cooled so it's not possible to conduct more tests at this point.

I must say that the demosaicing algorithm is very good to replicate almost same detail level. The true difference comes in narrow band where all the four matrix pixels will record signal same way.

Here is a small comparison of the same full spectrum target in true raw:...

 

[Herra Kuulapaa]

1 hour ago, GTom said:

That is quite a difference in intensity! Were they identically exposed? (time, aperture, target)

Target is exactly the same, but I don't remember how the exposure was so it's better to take this only as an example of spatial resolution.

In this sense it's hard for me to accept that the full monochrome signal would have a lower spatial resolution than RGB, especially with narrow band such Ha.

Edited November 19, 2016 by Herra Kuulapaa

[Gina]

I really can't see any treatment of a DSLR coming anywhere near the performance of the ZWO ASI1500MM-Cool.  There may have been improvements in DSLR technology but I can't see any reason for manufacturers of DSLRs to make them that much better for astro imaging, the scale of users would simply not warrant it.  And if such a camera were available it would be bound to have a similar cost to the ZWO camera.

When I was experimenting with debayered DSLRs I was comparing with CCD sensors at two or three times the cost of the ASI1800MM-Cool.  I certainly spent more than one of those ZWO cameras cost had they been available.

Edited November 19, 2016 by Gina

[GTom]

I think I see here, that you're completely right... (mono modded D600 vs ASI1600mm-cool). A few more pics and I am convinced... I started a smalltalk around the a7s vs the ZWO, the a7s might be the last stand of consumer cameras here. However, mono-modding it is a financial Russian roulette...

12 minutes ago, Gina said:

I really can't see any treatment of a DSLR coming anywhere near the performance of the ZWO ASI1500MM-Cool.  There may have been improvements in DSLR technology but I can't see any reason for manufacturers of DSLRs to make them that much better for astro imaging, the scale of users would simply not warrant it.  And if such a camera were available it would be bound to have a similar cost to the ZWO camera.

When I was experimenting with debayered DSLRs I was comparing with CCD sensors at two or three times the cost of the ASI1800MM-Cool.  I certainly spent more than one of those ZWO cameras cost had they been available.

[Adam J]

4 hours ago, Herra Kuulapaa said:

Hmmm., there is something funny in your test. It might be the sensor too, because Exmors seem to behave very nice after mono conversion. They have much more narrower conductors than 1000D though, which may decrease the amount of photons hitting them.

Here is some real world testing

Here is a comparison of full spectrum RGB D5100 on left vs D5100am (astromono) on right. Both files are true raw (non debayered).

click it large

The difference between your test and mine is more likely to do with the face that mine was conducted in low light through a H-a filter. I am guessing this is daylight with no filter. In which case the increased performance against a broad band source is to be expected.

The astro images you linked look good but I guess its not a side by side comparison of modded and non modded. Like I say the proof will be when I do that actual side by side. Until then I am not drawing any final conclusions.

Yes the 1000D has larger tracks however its pixels are also proportionally larger. Like I say as soon as we get a clear night I will know for sure.

Edited November 19, 2016 by Adam J

[Adam J]

1 hour ago, Gina said:

I really can't see any treatment of a DSLR coming anywhere near the performance of the ZWO ASI1500MM-Cool.  There may have been improvements in DSLR technology but I can't see any reason for manufacturers of DSLRs to make them that much better for astro imaging, the scale of users would simply not warrant it.  And if such a camera were available it would be bound to have a similar cost to the ZWO camera.

When I was experimenting with debayered DSLRs I was comparing with CCD sensors at two or three times the cost of the ASI1800MM-Cool.  I certainly spent more than one of those ZWO cameras cost had they been available.

I dont think anyone is going to argue that the ZWO ASI1600MM-Cool is not a superior sensor for astro work across the board and by quite a long way. I have personally only spent about 50 pounds doing this as I got a broken 1000D for 40 pounds fixed it and purchased some polish on top of that. I simply cant spend over 1000 pounds on an astro camera. If this mod is not going to give me better performance I will just stick my 450D into the cool box and use that without mono modding it. it actually takes a pretty decent image at -15c. 

 

I would not recommend anyone paying for a conversion as that would not be cost effective in comparison to a dedicated camera like the ASI, for the same reason I would not recommend that someone try to convert one of the more expensive DSLR's. But if you have the skill and a lack of cash then it might help is certain areas if you convert a budget DSLR. If only for lum for galaxies or reflection nebula. To be honest its also been fun doing it.

This mod really is for someone who cant afford cameras like the ASI not for someone trying to equal their performance for less cash because they are going to be disappointed.

Edited November 19, 2016 by Adam J

[GTom]

Everybody has a different budget, approach and goals in this hobby. Anyway, lots of satisfied people are using self-modded "color" DSLR-s budgeting them at their normal photo market value. I am actually reviving my astro-hobby after several years of "absence" and need a bit of a guidance.
I have a bit of a budget on an airline-portable setup now, but want to spend it wisely. 

For me, time is very precious, those tens of hours of exposures one often see on astrobin, would take years in my case even for circumpolar objects. Boosting camera efficiency by a factor of 2, not to mention more, would bring me much closer to the result. As I understood, B&W conversion
in theory (and from the real life samples) should bring 1EV advantage - this is like taking a 130mm APO instead of my current 90/600! Going forward I was curious if there is a meaningful second step, like 2+EV advantage by using a dedicated astro camera as the ZWO ASI.

Speaking of price, one doesn't necessarily need the "top dog" asi1600mm, there are solutions at 1/2price with 8mm diameter - which might be perfectly fine for smaller objects.

1 hour ago, Adam J said:

I dont think anyone is going to argue that the ZWO ASI1600MM-Cool is not a superior sensor for astro work across the board and by quite a long way. I have personally only spent about 50 pounds doing this as I got a broken 1000D for 40 pounds fixed it and purchased some polish on top of that. I simply cant spend over 1000 pounds on an astro camera. If this mod is not going to give me better performance I will just stick my 450D into the cool box and use that without mono modding it. it actually takes a pretty decent image at -15c. 

 

I would not recommend anyone paying for a conversion as that would not be cost effective in comparison to a dedicated camera like the ASI, for the same reason I would not recommend that someone try to convert one of the more expensive DSLR's. But if you have the skill and a lack of cash then it might help is certain areas if you convert a budget DSLR. If only for lum for galaxies or reflection nebula. To be honest its also been fun doing it.

This mod really is for someone who cant afford cameras like the ASI not for someone trying to equal their performance for less cash because they are going to be disappointed.

Edited November 19, 2016 by GTom

[russellhq]

13 hours ago, GTom said:

Everybody has a different budget, approach and goals in this hobby. Anyway, lots of satisfied people are using self-modded "color" DSLR-s budgeting them at their normal photo market value. I am actually reviving my astro-hobby after several years of "absence" and need a bit of a guidance.
I have a bit of a budget on an airline-portable setup now, but want to spend it wisely. 

For me, time is very precious, those tens of hours of exposures one often see on astrobin, would take years in my case even for circumpolar objects. Boosting camera efficiency by a factor of 2, not to mention more, would bring me much closer to the result. As I understood, B&W conversion
in theory (and from the real life samples) should bring 1EV advantage - this is like taking a 130mm APO instead of my current 90/600! Going forward I was curious if there is a meaningful second step, like 2+EV advantage by using a dedicated astro camera as the ZWO ASI.

Speaking of price, one doesn't necessarily need the "top dog" asi1600mm, there are solutions at 1/2price with 8mm diameter - which might be perfectly fine for smaller objects.

Some pretty impressive work has been achieved with a standard DSLR by Scott Rosen http://www.astronomersdoitinthedark.com

I think the key to Scott's success is dark skies: they make everything better ?

[GTom]

1 hour ago, russellhq said:

Some pretty impressive work has been achieved with a standard DSLR by Scott Rosen http://www.astronomersdoitinthedark.com

I think the key to Scott's success is dark skies: they make everything better ?

Of course, if someone has a Uluru or equivalent in the backyard ...

[Herra Kuulapaa]

21 hours ago, Adam J said:

The difference between your test and mine is more likely to do with the face that mine was conducted in low light through a H-a filter. I am guessing this is daylight with no filter. In which case the increased performance against a broad band source is to be expected.

The astro images you linked look good but I guess its not a side by side comparison of modded and non modded. Like I say the proof will be when I do that actual side by side. Until then I am not drawing any final conclusions.

Yes the 1000D has larger tracks however its pixels are also proportionally larger. Like I say as soon as we get a clear night I will know for sure.

Ah, I have forgotten much data during the years. Anyway, here is one comparison done with both full spectrum RGB and mono. Cameras were D5100 and the light was white led through Ha filter. Identical exposure by all of the variables and true sensor raw extracted.

 

And a quick n dirty histogram:

Far right are RGB red pixels. Second form the right is mono. This is the amount of signal gained (~8,5%) by micro lenses with IMX071 - but don't extrapolate this into any other sensor or even mono mod type. It's only valid with the process I used and polishing deeper or shallower will most likely result something else. Anyway, far on the left are the blue pixels and second from the left are green pixels. Remember, there is signal in all of the three color pixels despite of the red Ha color and RGB pixels have absorption as well - removing them will slightly compensate lost micro lenses.

Edit: Summary: Now I'm talking about IMX071 sensor (Nikon D5100). There will be increased spatial resolution because all of the mono pixels record the signal same way. This is seen with mono CCD vs RGB CCD comparisons as well. Mono mod will result a decrease in signal (lower QE), but the difference is not many fold like mentioned before and it's easily compensated with small increase of exposure time. Best option will still be a native mono CMOS astrocamera, but a DSLR monomod is a valid and possible path into narrow band imaging when the budget is tight and especially if a full frame sensor FOV is preferred.

Edited November 20, 2016 by Herra Kuulapaa