Debayering a DSLR's Bayer matrix.

[Gina]

This is the one I'm thinking of getting:

http://uk.farnell.com/jsp/search/productdetail.jsp?sku=1639752

It's 2A at 15.7V. dT 75C

The next one up is 3.9A at 15.7V. dT 75C: that might be too much power.

I think that might have been the one I bought (well bought 2 actually).  I think it was a 2A one.

EDIT ---  Found it - and yes, it was MCPE-127-10-25   Works very well though I forget what temperature I got down to.  It will be in one of my cooling threads   I'll have the answer again soon though when I put an 1100D in the cooler instead of the 450D,

Edited April 25, 2014 by Gina

[russellhq]

I've just checked the figures my Arduino sketch is providing with an online calculator for dewpoint and the calculator shows about -25C whereas my sketch is showing -11 for temp of 18C and RH of 13    I got the DP calculation sketch from the Arduino web site!

Hmmm.  This calc gives a dew point of -11C which agrees with the sketch.

http://www.daytonashrae.org/psychrometrics_si.html

Also, looking at various charts, -11C looks to be much closer to the ball park than -25C I'm afraid.

[libraryman]

I think that might have been the one I bought (well bought 2 actually).  I think it was a 2A one.

EDIT ---  Found it - and yes, it was MCPE-127-10-25   Works very well though I forget what temperature I got down to.  It will be in one of my cooling threads   I'll have the answer again soon though when I put an 1100D in the cooler instead of the 450D,

Excellent Gina, thank you

Ray

[Gina]

Hmmm.  This calc gives a dew point of -11C which agrees with the sketch.

http://www.daytonashrae.org/psychrometrics_si.html

Also, looking at various charts, -11C looks to be much closer to the ball park than -25C I'm afraid.

Good - the sketch is alright then   Anyway, I shall be testing with low temperature so we shall see how good it is.  I have virtually no doubt about achieving -10C without misting up and that's fine

[russellhq]

Has anyone noticed that the 1100D signal levels seem to drop once debayered?

I debayered half the sensor and took a flat then measured the pixel values on either side. The debayered side was on average lower.

Here's an example of the data I recorded:

Debayered Results

Pixel 1: 1246

Pixel 2: 1265

Pixel 3: 1231

Pixel 4: 1292

Average: 1258

Original Results

R: 786

G: 1442

G: 1444

B: 1491

Average: 1291

Here's a copy of the flat:

https://www.dropbox.com/s/uyz9zm6v7v0aj1g/IMG_5489.CR2

[StuartJPP]

Doesn't debayering also remove the microlenses?

I think the Canon 1100D has microlenses though not 100% sure.

[Gina]

Yes, it has.  And yes, the sensitivity is reduced but every pixel will respond to all colours.  I thought the reduction was more than that   Nice to have actual figures.

Edited April 26, 2014 by Gina

[StuartJPP]

I suppose that you lose something by removing the microlenses and gain something by removing the Bayer filters.

Overall you lose more from the removal of the microlenses though.

Imagine being able to remove the Bayer matrix without losing the microlenses.... some sort of laser

[Gina]

Been sorting out 1100Ds.  That non-working one I bought recently works fine with external power   The one with the cold finger had a duff sensor in it - one I had been debayering.  There was a broken gold wire where my debayering managed to creep under the filter frame (that I was using as a guard) and catch a wire.  I did test it  - with short exposures up to 1/50th sec the image was black and at 1s and more in daylight, all white.  Usually after destroying by debayering the sensor won't work at all. 

The sensor assembly that was untouched (without even filter removal) works fine and I have taken the shield off, snipped off a lug that was in the way and slipped a cold finger in together with plastic insulator to prevent shorting on the SMD components on the back of the image processor board.  That's now ready for cooling test.  I have tested it with the cold finger and it's still working.  Now I need to look out a DS18B20 to measure the CF temperature and rig up a cooling circuit.

I've made further progress on the cover glass replacement rig, sorting out a design for a printed plastic frame to hold the sensor.  Now I know which cold finger I'm using I can finalise the design.  I have the 3D printer working well enough to do the printing

[sharkmelley]

Has anyone noticed that the 1100D signal levels seem to drop once debayered?

I debayered half the sensor and took a flat then measured the pixel values on either side. The debayered side was on average lower.

That's a very interesting result.   Intuitively (as someone who has never tried debayering)  I would expect each debayered pixel to have a value equal to the sum of R,G & B values from the CFA pixels and then reduced by a factor due to the loss of the microlenses.  So are we saying that the loss of the microlenses reduces the sensitivity by a factor of around 3 or so?  Or is something else happening as well?

If this result is typical then it indicates to me that debayering an 1100D will result in very little overall advantage from an imaging point of view, though it's a very interesting technical exercise.

Is this result also typical of those obtained from other sensor types?

Thanks for posting the CR2 file, it was most informative

Mark

[Gina]

The sensitivity to Ha is definitely increased, presumably because the spectral response of CFA red pixels falls off in the deep red.  Likewise the sensitivity to OIII is increased because the spectral line of OIII falls between the green and blue responses.  SII is even deeper in the red than Ha.  So a debayered DSLR achieves better sensitivity for NB as well as having twice the resolution in both axes.  For RGB the sensitivity may be slightly lower but the resolution is still doubled.  This is definitely a very worthwhile mod for astro use.  I have yet to compare a debayered 1100D directly with a mono CCD astro camera but would still expect the CCD camera to win, particularly with regard to noise.  But the cost of a 12mp astro camera is considerable.  Even with the cost of all the DSLRs I've bought for this project it is still only a small proportion of the cost of a 12mp astro CCD camera.  But also, of course, I enjoy the experimenting

Edited April 28, 2014 by Gina

[russellhq]

That's a very interesting result.   Intuitively (as someone who has never tried debayering)  I would expect each debayered pixel to have a value equal to the sum of R,G & B values from the CFA pixels and then reduced by a factor due to the loss of the microlenses.  So are we saying that the loss of the microlenses reduces the sensitivity by a factor of around 3 or so?  Or is something else happening as well?

If this result is typical then it indicates to me that debayering an 1100D will result in very little overall advantage from an imaging point of view, though it's a very interesting technical exercise.

Is this result also typical of those obtained from other sensor types?

Thanks for posting the CR2 file, it was most informative

Mark

I'm not sure it works this way. A good experiment would be to compare a mono ccd luminance Flat frame with mono ccd R, G & B filtered Flat Frame. And see if the ADUs add up?

[sharkmelley]

I'm not sure it works this way. A good experiment would be to compare a mono ccd luminance Flat frame with mono ccd R, G & B filtered Flat Frame. And see if the ADUs add up?

The flaw in my argument is that the flat frame needs to be bias subtracted. 

I'm assuming the CR2 file you posted was recorded using white light (since the R,G & B pixels all have values) The debayered pixels are now able to collect the R,G & B elements of the light because the filter has been removed.  So once the bias is subtracted then the (bias-subtracted) pixels on the left hand side (the debayered side) ought to show roughly the sum of the (bias-subtracted) R,G,B on the right hand side (the non-debayered side) unless losing the microlenses is having a very significant effect.  

Does anyone know the bias level for the 1100D?  For other Canon sensors it is generally a power of 2 e.g. 512, 1024, 2048.    Armed with the bias level, a very rough approximation can be made of the effect of losing the microlenses.

Mark

[russellhq]

The flaw in my argument is that the flat frame needs to be bias subtracted. 

I'm assuming the CR2 file you posted was recorded using white light (since the R,G & B pixels all have values) The debayered pixels are now able to collect the R,G & B elements of the light because the filter has been removed.  So once the bias is subtracted then the (bias-subtracted) pixels on the left hand side (the debayered side) ought to show roughly the sum of the (bias-subtracted) R,G,B on the right hand side (the non-debayered side) unless losing the microlenses is having a very significant effect.  

Does anyone know the bias level for the 1100D?  For other Canon sensors it is generally a power of 2 e.g. 512, 1024, 2048.    Armed with the bias level, a very rough approximation can be made of the effect of losing the microlenses.

Mark

I'm pretty sure the black level on the 1100D is 2046-2048, but this was already subtracted from the values I posted

[pixueto]

I think the main reason for having a debayered DSLR is narrowband imaging. Now you have every pixel collecting light, say with a H alpha as oposed to only 25% of them. This is a huge improvement. For colour images, I think we would be better off with a sensor with the CFA intact as the colour interpolation algorithms are pretty good and the microfilters are still there. But for narrowband imaging... definitely, the mod is truly worthwhile.

Edited April 28, 2014 by pixueto

[Gina]

I'm still of two minds regarding using a debayered DSLR for LRGB.  I'm definitely going for NB as the main reason for CFA removal but I'm weighing up what I would image.  My experience with the transparency (or lack of it) and seeing we have in this country, the image resolution is limited by the sky and there is no point in going for small pixels for longer focal lengths.  OK the DSLR image size will cover a greater FOV but AFAICT there are few DSOs warranting this frame size and LRGB.  In fact the only DSOs I use LRGB for are galaxies and maybe adding star colour to NB.  In view of the number of 1100Ds I seem to have I'm thinking I might keep one for use as an OSC camera (assuming I don't destroy all the 1100D sensors I have in total).  This would save me buying a set of 36mm unmounted LRGB filters and at the same time mean I would need only a 4 position filter wheel on the debayered camera.  The result would be a smaller and sturdier all singing, all dancing camera for narrow band imaging.

Edited April 28, 2014 by Gina

[russellhq]

I'm still of two minds regarding using a debayered DSLR for LRGB. I'm definitely going for NB as the main reason for CFA removal but I'm weighing up what I would image. My experience with the transparency (or lack of it) and seeing we have in this country, the image resolution is limited by the sky and there is no point in going for small pixels for longer focal lengths. OK the DSLR image size will cover a greater FOV but AFAICT there are few DSOs warranting this frame size and LRGB. In fact the only DSOs I use LRGB for are galaxies and maybe adding star colour to NB. In view of the number of 1100Ds I seem to have I'm thinking I might keep one for use as an OSC camera (assuming I don't destroy all the 1100D sensors I have in total). This would save me buying a set of 36mm unmounted LRGB filters and at the same time mean I would need only a 4 position filter wheel on the debayered camera. The result would be a smaller and sturdier all singing, all dancing camera for narrow band imaging.

I was considering the possibility of debayering half the sensor for imaging galaxies. Half the sensor should be plenty big enough and you have a camera that will capture luminance and rgb just by shifting the scope slightly.

No idea if it would work though!

[pixueto]

Any luck guys with the dremel and the felt buffing tool dry? I got to a point where I can remove the CFA entirely but there are some sort of coating (coating, no  CFA leftovers) left in places. I can remove this applying the dremel with the felt tips for longer but I am concerned that I may end up damaging the sensor by doing so. I also thought about applying the rotary tool and the felt buffing head with fluid (isopropanol) when removing the coating to reduce strain in the sensor.

Does anybody have an idea what this coating is made of? Because of it, the flat frames aren't perfect but it's a pity as the bayer matrix is no longer there. It goes away very easily with the felt buffing tip. I can't help thinking there must be some safe solvent out there capable or removing this thin layer without subjecting the sensor to more abrasion. Any ideas?

Edited April 28, 2014 by pixueto

[Gina]

I was considering the possibility of debayering half the sensor for imaging galaxies. Half the sensor should be plenty big enough and you have a camera that will capture luminance and rgb just by shifting the scope slightly.

No idea if it would work though!

That's radical   I think it would be difficult to get a clean changeover and also half sensor size partly negates the benefit of a DSLR with it's relatively large image sensor.

[pixueto]

Any luck guys with the dremel and the felt buffing tool dry? I got to a point where I can remove the CFA entirely but there are some sort of coating (coating, no  CFA leftovers) left in places. I can remove this applying the dremel with the felt tips for longer but I am concerned that I may end up damaging the sensor by doing so. I also thought about applying the rotary tool and the felt buffing head with fluid (isopropanol) when removing the coating to reduce strain in the sensor.

Does anybody have an idea what this coating is made of? Because of it, the flat frames aren't perfect but it's a pity as the bayer matrix is no longer there. It goes away very easily with the felt buffing tip. I can't help thinking there must be some safe solvent out there capable or removing this thin layer without subjecting the sensor to more abrasion. Any ideas? Could it be a bottom antireflection coating color filter? I have tried nail polish but it doesn't go away without the dremel.

Edited April 28, 2014 by pixueto

[Gina]

Any luck guys with the dremel and the felt buffing tool dry? I got to a point where I can remove the CFA entirely but there are some sort of coating (coating, no  CFA leftovers) left in places. I can remove this applying the dremel with the felt tips for longer but I am concerned that I may end up damaging the sensor by doing so. I also thought about applying the rotary tool and the felt buffing head with fluid (isopropanol) when removing the coating to reduce strain in the sensor.

Does anybody have an idea what this coating is made of? Because of it, the flat frames aren't perfect but it's a pity as the bayer matrix is no longer there. It goes away very easily with the felt buffing tip. I can't help thinking there must be some safe solvent out there capable or removing this thin layer without subjecting the sensor to more abrasion. Any ideas?

I haven't got round to trying the buffing tool dry as yet and I have no idea how to remove that layer below the CFA.  I shall endeavour to get the sensor as flat as I can, hoping not to go too far.  I think that if I can remove all the CFA over most of the image area, I shall make do with that and use flats to clean up the final image. 

I've been trying to get my dreaded 3D printer to produce usable prints!  I'm up to the 17th attempt at a reasonable frame for holding the sensor for drying and replacing the cover glass and the 4th attempt at the lever that holds the glass and applies it to the sensor frame    I'm daft really - in the time I've spent on it I could have made one from sheet plastic or wood with hand tools and hot melt glue!   And in fact, thinking about it, I may give up and make it by hand!  I'm sure it would be far less frustrating.

[sharkmelley]

I'm pretty sure the black level on the 1100D is 2046-2048, but this was already subtracted from the values I posted

So my original analysis stands then.  Removing the microlenses (and maybe some additional effect) is reducing the pixel sensitivity by a factor of very roughly 3. 

I think the main reason for having a debayered DSLR is narrowband imaging. Now you have every pixel collecting light, say with a H alpha as oposed to only 25% of them. This is a huge improvement. For colour images, I think we would be better off with a sensor with the CFA intact as the colour interpolation algorithms are pretty good and the microfilters are still there. But for narrowband imaging... definitely, the mod is truly worthwhile.

This has always been my understanding also.  All the pixels will now be sensitive to H-alpha instead of one quarter of them.  But if the price to pay is a reduction in sensitivity by a factor of 3 then not much has been gained. 

I'd love to be proved wrong so if someone has some evidence of the improvement.  I would also happily accept that resultswith  the 1100D may not be typical of what can be achieved with debayering other DSLR sensors.

Mark

[Gina]

I may have those comparisons one day

[pixueto]

Found more information about the bottom antireflection coating colour filter:

In general, BARC layer 16 may be formed from any conventional BARC material, including a dyed organic film-forming BARC material or a light-absorbing polymeric film-forming BARC material. In some embodiments, BARC layer 16 preferably is substantially absorptive of radiation in the wavelength range used to pattern color filter array 14 and is substantially transmissive to radiation in the wavelength range to be imaged by image sensor system 10 (e.g., the visible radiation spectrum). BARC layer 16 may be formed from an organic film-forming material or a polymeric film-forming material. In one embodiment, BARC layer 16 is a photoresist-based antireflective coating that is substantially transmissive to radiation in a wavelength range of about 400 nm to about 700 nm (e.g., a Shipley AR2-600 antireflection coating, which is available from Shipley Company, L.L.C. of Marlborough, Mass., U.S.A.). In this embodiment, BARC layer 16 may be applied by a conventional spin-coater operating at 2000 rpm during deposition and at 4790 rpm during spreading; the resulting BARC layer 16 has a thickness of about 60 nm. After deposition, BARC layer 16 is exposed to a 60 second proximity bake at 205° C. on a DNS track.

Any ideas which solvent can be used to remove it?

[russellhq]

Here's one method :

http://www.google.com/patents/US8530356

Plus side, the gold wires won't react with the nitric acid.