Debayering a DSLR's Bayer matrix.

[Dave_D]

"Even going gently (best results are obtained by scraping with a small baguette)" that made me laugh...

[GTom]

On 10/24/2016 at 22:34, Adam J said:

Interesting finding Adam! Are those areas relatively close together? Lens vignetting might also cause hiccups... Total values are convincing though, it appears that you get almost a full EV - in other words, you can get the same illumination at a given wavelength in 40% less time.

 

I found spare sensors on ebay for my sony A6000, I just wonder is it possible to check if a debayered sensor is fine without putting it into a camera? That would save the hassle to disassemble-reassemble every time...

Edited November 16, 2016 by GTom

[GTom]

BTW, other than ebay, is there any source for a replacement a6000 sensor?

[Adam J]

I dont think that anyone has ever done an A6000...well not that I have read about. It would be a jump into the unknown. I am not even really familiar with the model. What I have heard is that the sensors do get more delicate as the pixel count increases, I would not do this to an expensive sensor first time out.

I am going to post a real world comparison soon. I have two identical 1000D's one Mono one not. I am getting some very very confusing results from them. Under some exposure conditions the mono looks marginally better, under other conditions the RGB is hugely superior despite only using 1/4 of the pixels. I have decided that the only way to know for sure is to point each at the California nebula for 10 x 5 min exposures each and see how the two images compare. I JUST NEED SOME CLEAR SKY I HAVE HAD THE MONO FOR OVER A MONTH NOW!!

 

[GTom]

One more vote for clear sky, its on extremely short supply on the northern hemisphere now:(((.

At first I didn't want to "fix" my a6000, but when I saw that 70$ buys me a sensor on ebay I gave a thought. I'll definitely wait until your results, if you have the chance, do o-iii or h-beta as well.

So far I've seen many people scraping off their Bayer's, but very little comparison. Your post showed the first quantitative results...

 

Just an addition: the a6000 must be possible, Maxmax sells monochrome a6000's for 2k$, and it wouldn't be the first "NEX" to be converted. The not-so-extreme pricing indicates that the process shouldn't be extremely difficult...

This blog seems to be sleeping, but the downloadable price list shows a whole lot of cameras, from all brands.

A little French reading, finally the quantitative astro comparison I was seeking...

Edited November 17, 2016 by GTom

[Adam J]

10 hours ago, GTom said:

One more vote for clear sky, its on extremely short supply on the northern hemisphere now:(((.

At first I didn't want to "fix" my a6000, but when I saw that 70$ buys me a sensor on ebay I gave a thought. I'll definitely wait until your results, if you have the chance, do o-iii or h-beta as well.

So far I've seen many people scraping off their Bayer's, but very little comparison. Your post showed the first quantitative results...

 

Just an addition: the a6000 must be possible, Maxmax sells monochrome a6000's for 2k$, and it wouldn't be the first "NEX" to be converted. The not-so-extreme pricing indicates that the process shouldn't be extremely difficult...

This blog seems to be sleeping, but the downloadable price list shows a whole lot of cameras, from all brands.

A little French reading, finally the quantitative astro comparison I was seeking...

There have been other quantitative studies, however for the most part people have done these with partially removed coatings rather then by a comparison between two camera with and without. Essentially what I am seeing is that she very short daylight exposures (1 second or so) through the H-A filter the RGB has the clear advantage. However, once I move to longer exposures in a dark room the results reverse and the mono is superior. 

 

What I have seen though is that the mono will provide a significant performance boost when Luminance is shot through my CLS filter. Kinda makes sense as all the pixels are sensitive to all wavelengths and so each pixel is receiving 3 to 4 times the amount of light as it would normally against a broadband target like a Galaxy or a reflection nebula. This is counter to what other people have suggested...everything I have read so far indicates that people think the biggest boost will be when using narrow band, but I do believe this to be completely correct due to the loss of per pixel sensitivity.  

Edited November 17, 2016 by Adam J

[GTom]

1 hour ago, Adam J said:

There have been other quantitative studies, however for the most part people have done these with partially removed coatings rather then by a comparison between two camera with and without. Essentially what I am seeing is that she very short daylight exposures (1 second or so) through the H-A filter the RGB has the clear advantage. However, once I move to longer exposures in a dark room the results reverse and the mono is superior. 

Wow, this is a very weird behavior... I was expecting a lot of gain on the red channel - since it got only a single pixel out of 4 in the Bayer matrix... One thing, did you force the WB on a set value (e.g. 6000K)?

Quote

What I have seen though is that the mono will provide a significant performance boost when Luminance is shot through my CLS filter. Kinda makes sense as all the pixels are sensitive to all wavelengths and so each pixel is receiving 3 to 4 times the amount of light as it would normally against a broadband target like a Galaxy or a reflection nebula. This is counter to what other people have suggested...everything I have read so far indicates that people thing the biggest boost will be when using narrow band, but I do believe this to be completely correct due to the loss of per pixel sensitivity.  

CLS through Bayer means that each subpixel looses ~60% or 1+EV, not much surprised an advancement even in green. Here is an estimate, that microlenses bring 40% to the equation.

Edited November 17, 2016 by GTom

[Adam J]

33 minutes ago, GTom said:

Wow, this is a very weird behavior... I was expecting a lot of gain on the red channel - since it got only a single pixel out of 4 in the Bayer matrix...

CLS through Bayer means that each subpixel looses ~60% or 1+EV, not much surprised an advancement even in green. Here is an estimate, that microlenses bring 40% to the equation.

Well this is part of the issue, you cant do a comparison based on short daylight exposures only as its only part of the story. You don't get much thermal noise in short exposure, only the read noise....although this is also the case in a 10 min exposure when the camera is cooled.

The other interesting thing is that the effectiveness of the micro-lenses will also be dependent on the angle of incidence of the light against the CMOS sensor at least for conventional front illuminated sensors. Back illuminated sensors would not see that effect. So part of my issue is that I am using a fast 50mm lens to do my experiments and so the light cone will be very tight, meaning what with no micro lens allot more light will miss the light sensitive area and hit the copper...or at least that is my theory. There is of course a range of angles in any given focal point however the point is that the spread is higher for  short focal lengths leading to greater light loss without micro-lenses. Again at least in theory...

 

  

[GTom]

Correct, I probably have to forget my nice & shiny 1.4/20 Sigma ART lens with the debayered body

[Adam J]

46 minutes ago, GTom said:

Correct, I probably have to forget my nice & shiny 1.4/20 Sigma ART lens with the debayered body

Probably, but it will be dependent on the ratio of the height of the pixel to its width and although most diagrams I have seen show this to be one to one I am not sure if this is a true representation of what is happening on the chip itself. 

[GTom]

In more recent sensors the "eye" doesn't sit too deep AFAIK, even if they are not BSI. I recall Panasonic claiming that they moved the sensing part upwards in their m4/3 sensor few years ago (canon 3xx was history even that time).

[Xplode]

3 hours ago, GTom said:

Wow, this is a very weird behavior... I was expecting a lot of gain on the red channel - since it got only a single pixel out of 4 in the Bayer matrix... One thing, did you force the WB on a set value (e.g. 6000K)?

CLS through Bayer means that each subpixel looses ~60% or 1+EV, not much surprised an advancement even in green. Here is an estimate, that microlenses bring 40% to the equation.

It is not weird behavior at all because when you take Ha with a mono DSLR you gain resolution, not neccesarily signal as most people think.

Where you will gain signal is in the stacked image because what many people do wrong with DSLR Ha data is they don't remove the green and blue channels before stacking the images meaning they interpolate 1 pixel with signal with 3x pixels with just noise, that's not gonna make the result have a good signal...

 

What i do find weird is that there is a big loss in Ha sensitivity, but a small gain in OIII sensitivity.

 

Like Adam J wrote the real gain is in being able to capture real luminance and it's something i have done with great success with my 6Dm.

I have also done true LRGB imaging with color filters for all channels and just like for a CCD camera there isn't much RGB needed for the image to be pretty good when it's combined with luminance.

 

Here's M33 HaLRGB with no noise reduction at all, it's almost 15 hours total with 1 hour of R, G and B, 4 hours Ha and 8 hours luminance.

[GTom]

7 minutes ago, Xplode said:

It is not weird behavior at all because when you take Ha with a mono DSLR you gain resolution, not neccesarily signal as most people think.

Where you will gain signal is in the stacked image because what many people do wrong with DSLR Ha data is they don't remove the green and blue channels before stacking the images meaning they interpolate 1 pixel with signal with 3x pixels with just noise, that's not gonna make the result have a good signal...

 

What i do find weird is that there is a big loss in Ha sensitivity, but a small gain in OIII sensitivity.

 

Like Adam J wrote the real gain is in being able to capture real luminance and it's something i have done with great success with my 6Dm.

I have also done true LRGB imaging with color filters for all channels and just like for a CCD camera there isn't much RGB needed for the image to be pretty good when it's combined with luminance.

 

Here's M33 HaLRGB with no noise reduction at all, it's almost 15 hours total with 1 hour of R, G and B, 4 hours Ha and 8 hours luminance.

...

Nice shot, especially from a 100mm APO! Added "Resolution" you call it, is actually additional information of the same quality presented now by 4 pixels instead of one. If I understand correctly, a 4:1 median filtering on the monochrome material, you get the same resolution as with the Bayer but significantly less noise. Or the same noise level but higher detail.

[Xplode]

It's taken with a 12" newtonian, the 100mm have not been in much use since i bought the newtonian.

The 100mm is back on the mount now thou...but the mount is down due to something wrong with the electronics

 

Sounds about right except it would be 2:1 meidan filtering, not 4:1 (4 would mean an area of 4x4, 2 = 2x2)

[Adam J]

2 hours ago, Xplode said:

It is not weird behavior at all because when you take Ha with a mono DSLR you gain resolution, not neccesarily signal as most people think.

Where you will gain signal is in the stacked image because what many people do wrong with DSLR Ha data is they don't remove the green and blue channels before stacking the images meaning they interpolate 1 pixel with signal with 3x pixels with just noise, that's not gonna make the result have a good signal...

 

What i do find weird is that there is a big loss in Ha sensitivity, but a small gain in OIII sensitivity.

 

Like Adam J wrote the real gain is in being able to capture real luminance and it's something i have done with great success with my 6Dm.

I have also done true LRGB imaging with color filters for all channels and just like for a CCD camera there isn't much RGB needed for the image to be pretty good when it's combined with luminance.

 

Here's M33 HaLRGB with no noise reduction at all, it's almost 15 hours total with 1 hour of R, G and B, 4 hours Ha and 8 hours luminance.

Yes I think this is the real advantage is (luminance). Have you ever done a Mono vs RBG comparison with a H-A filter? Also do you cool your camera?

I always remove the green and blue channels

In  so far as those results are concerned I am no longer sure that they are correct, mostly because while those results where using different parts of the same sensor with the perimeter for the non mono readings the readings for the second normal RGB camera do not yield the same results. I am going to fully wright this all up once I have some real world H-A results.

Edited November 17, 2016 by Adam J

[GTom]

2 hours ago, Adam J said:

Yes I think this is the real advantage is (luminance). Have you ever done a Mono vs RBG comparison with a H-A filter? Also do you cool your camera?

I always remove the green and blue channels

In  so far as those results are concerned I am no longer sure that they are correct, mostly because while those results where using different parts of the same sensor with the perimeter for the non mono readings the readings for the second normal RGB camera do not yield the same results. I am going to fully wright this all up once I have some real world H-A results.

We fully appreciate your efforts! Yes, it is difficult to compare...

WB and metering are the first candidates to go mad because of the different conditions: you need to select a fixed CT and shoot full manual.

I'd also be curious if a B&W converted, cooled APSc camera can come close to astronomical CCD/CMOS cams, because an  ASI 1600MM-Cool can be had for ~1600EUR (+few KAF8300 candidates join in if one considers 2nd hand). If I take the B&W conversion road, I most likely grab a commercial conversion service like JTW.

However this case price is a question: the camera value + conversion costs (+hassle...) should be definitely below the dedicated solution...

Edited November 17, 2016 by GTom

[Adam J]

If  you have the cash get the ASI i am only doing this because my total sensor budget is under 200 pounds.

[Stub Mandrel]

5 hours ago, Xplode said:

Where you will gain signal is in the stacked image because what many people do wrong with DSLR Ha data is they don't remove the green and blue channels before stacking the images meaning they interpolate 1 pixel with signal with 3x pixels with just noise, that's not gonna make the result have a good signal...

Good tip

I have experimented with stacking DSLR data with two  synthetic luminance layers, one from RGB and one from just R.

[Xplode]

2 hours ago, Adam J said:

Yes I think this is the real advantage is (luminance). Have you ever done a Mono vs RBG comparison with a H-A filter? Also do you cool your camera?

I always remove the green and blue channels

In  so far as those results are concerned I am no longer sure that they are correct, mostly because while those results where using different parts of the same sensor with the perimeter for the non mono readings the readings for the second normal RGB camera do not yield the same results. I am going to fully wright this all up once I have some real world H-A results.

Just natural cooling.

I unfortunately don't have Ha data taken with the same scope with both cameras

 

8 minutes ago, GTom said:

We fully appreciate your efforts! Yes, it is difficult to compare...

WB and metering are the first candidates to go mad because of the different conditions: you need to select a fixed CT and shoot full manual.

I'd also be curious if a B&W converted, cooled APSc camera can come close to astronomical CCD/CMOS cams, because an  ASI 1600MM-Cool can be had for ~1600EUR (+few KAF8300 candidates join in if one considers 2nd hand). If I take the B&W conversion road, I most likely grab a commercial conversion service like JTW.

However this case price is a question: the camera value + conversion costs (+hassle...) should be definitely below the dedicated solution...

WB should not matter since it's not used for RAW if converted through something like Pixsinsight unless it's actually enabled.

Testing should definitely be done full manual and also with max a few minutes apart so 2 cameras are needed or a sensor which is partly debayered.

A mono APS-C DSLR will come close to the 8300 even uncooled if the ambient temp isn't too high.

Newer DSLR's have very low noise at below 10C and doesn't have that hugh difference from cooling that ancient technology like CCD does.

Remember the 8300 sensor is over 10 years old technology. The only real hugh improvement in CCD technology the last years have been done by Sony. (also by other manufacturers but they make sensors we can't afford)

[GTom]

True, the venerable kaf8300 rather belongs to a museum than to a modern astrophotographer's setup. An ASI1600mm vs monochrome 5DmkII would be interesting though. The ASI has a smaller CMOS sensor but I guess with microlenses on place and the cooling is quite powerful. A mono-modded 5D would miss the microlens array and there is no practical way to cool it as efficiently.

[Xplode]

5D II? Why that camera?

It's performance for AP is way below the 6D, 5D III or newer cameras like the 7D II and 5D IV.

[GTom]

6 hours ago, Xplode said:

5D II? Why that camera?

It's performance for AP is way below the 6D, 5D III or newer cameras like the 7D II and 5D IV.

Not being in the canon camp I didn't know, just picked a cheaper Canon for the sake of a comparison. Let it be any recent camera... Question remains, how a decent DSLR/ILC stands up against the cheapest 4/3 factor astro-camera?

[Adam J]

12 hours ago, Xplode said:

Just natural cooling.

I unfortunately don't have Ha data taken with the same scope with both cameras

 

WB should not matter since it's not used for RAW if converted through something like Pixsinsight unless it's actually enabled.

Testing should definitely be done full manual and also with max a few minutes apart so 2 cameras are needed or a sensor which is partly debayered.

A mono APS-C DSLR will come close to the 8300 even uncooled if the ambient temp isn't too high.

Newer DSLR's have very low noise at below 10C and doesn't have that hugh difference from cooling that ancient technology like CCD does.

Remember the 8300 sensor is over 10 years old technology. The only real hugh improvement in CCD technology the last years have been done by Sony. (also by other manufacturers but they make sensors we can't afford)

I disagree I have seen the effect of white balance on raw images especially when processed as full resolution mono. However, I don't have pixsinaight so don't know what that work stream would involve.  I agree that CCD cameras are now on borrowed time but I am not so sure about a mono dslr coming close to a 8300 due to the loss of the micro lenses. Perhapse you could post some recent example images of weak h-a nebula with your exposure settings? 

Last night I set the camera up and just got a single 3 min frame of the soul nebula with the mono 1000D (I did not have time for the cooling). Assuming that I was pointing in the right place (ill plate solve it tonight)....but no real reason why not, then I got no indication of the nebula at all through my Baader 7nm H-a.  Although I would have done a 15 min cooled exposures for that target if cloud cover had allowed me, i might have expected a wisp of nebulosity despite sub optimal conditions.  

Edited November 18, 2016 by Adam J

[Xplode]

White balance has no effect on RAW unkless it's used when converting the image the RAW image.

Pixinsight has the option to disable or enable white balance , so does some of the other software used for converting RAW files.

Here is 8x600s with the 12" of the elephants trunk nebula.

Single 300s Ha image of the wizard nebula, unprocessed except it's been stretched a little. It has not been calibrated either.

 

Here's my settings in PI. It comes out as a greyscale image so it's treated just like a mono image from a CCD camera.

 

 

 

 

 

[GTom]

Very nice pics Ole, however 12" f/3.8 would probably yield yaw-dropping results with Bayer-on too - as a matter of fact I'd also check in with a roll of iso400 Provia :D. 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)?

Edited November 18, 2016 by GTom