Debayering a DSLR's Bayer matrix.

[russellhq]

I just tried some paint stripper on a webcam and it cleaned the bayer layer off perfectly in seconds and left a perfectly clean surface. I fell the canon sensor won't be so willing to play ball. Good news for webcam imagers though. Now i need a dead camera to test it with or i might do a little test on the sensor i have alread butchered but just do it in a corner.

I might try this on my webcam guide cam!

I just tried white spirit and no joy. What kind of paint stripper did you use?

[RAC]

I just tried white spirit and no joy. What kind of paint stripper did you use?

This. I spryed some on the bench then put just a small bit on the sensor. You want to keep this stuff away from any gold, or solder as it well eat it and make a big mess, not right away but give it a few days so clean it with meth or something afterwards.

[Gina]

Been trying to clean up my other spare sensor. That's the one I covered in electrically insulating thermal grease - a right mess. However, even after making really sure the connector for the ribbon cable was thoroughly clean, this sensor is giving Err 70. All I know is that it used to work but now it doesn't. So I haven't got a spare working sensor to use for testing. It's no use doing anything with a non-working sensor other than what I've already done with the other one. If I were to attack a sensor from a working camera the chances of getting into the sensor without destroying it in the process are pretty poor - certainly far worse than 50:50.

I am highly reluctant to give up but with the way the cover glass is fixed onto the sensor of the 1100D I think I have met my match There would be no point in getting a 1000D as this camera is already 2 stops less sensitive than the 1100D in addition to having a higher noise level. The other mods to the 1100D have been very successful so I guess I'll have to be satisfied with that. Debayering an 1100D sensor is simply an extreme mod too far - at least for me. Oh well, I put my best efforts into it.

[RAC]

I am highly reluctant to give up but with the way the cover glass is fixed onto the sensor of the 1100D I think I have met my match There would be no point in getting a 1000D as this camera is already 2 stops less sensitive than the 1100D in addition to having a higher noise level. The other mods to the 1100D have been very successful so I guess I'll have to be satisfied with that. Debayering an 1100D sensor is simply an extreme mod too far - at least for me. Oh well, I put my best efforts into it.

Its a bit of a [removed word] they changed so much about the way they made the later sensor. Atleast you can say you tried really hard.

Don't get too rapped up with the higher ISO levels of the 1100d over the 1000d, its only a gain and will do nothing to add photons, if i had one i would still be using iso400-1600 at most. The noise is only a little better but in the scheme of things makes little difference when both are cooled and you stack images. The best thing about the 1100d is its increase in bits from 12-14. I have been very happy with the performance of my 1000d's.

[Gina]

I notice the firm doing mono Canon DSLRs don't include the 1100D (Rebel T3) :- http://www.maxmax.com/monochrome_camera_order.asp

[Photosbykev]

I notice the firm doing mono Canon DSLRs don't include the 1100D (Rebel T3) :- http://www.maxmax.co...amera_order.asp

I noticed that as well Also on the Honis site the 550D has different sensor filter hardware, it seems a lot easier to remove the standard AA and IR filters

[Gina]

Its a bit of a [removed word] they changed so much about the way they made the later sensor. Atleast you can say you tried really hard.

True.

Don't get too rapped up with the higher ISO levels of the 1100d over the 1000d, its only a gain and will do nothing to add photons, if i had one i would still be using iso400-1600 at most. The noise is only a little better but in the scheme of things makes little difference when both are cooled and you stack images. The best thing about the 1100d is its increase in bits from 12-14. I have been very happy with the performance of my 1000d's.

I do most of my imaging at ISO 3200 but recent noise tests have shown that with cooling using ISO 6400 is well on.

[nytecam]

Very good project! So when they say the camera is 15 mega pixels, it's the raw pixels they are refering too, and so the debayered colour result is really only 3.75 mega-pixels (and not 15 mpix) after interpolation. Tut tut. So removing the CFA (the bayer layer) will increase your picture resolution to it's full 15 mpix (for a 15-mpix camera) grey scale, not that we can ever attain that kind of resolution here in the UK with our skies.

So you're saying a 15Mp colour cam is only 3.75Mp in resolution - don't think the maths works like that - if so then we've all been conned Surely the camera sensor uses two elements eg the luminance provides the image detail and dynamic range and chrominance adds blurred colour 'wash' which the eye find fully acceptable. Ok the bayered resolution can't be 100% but neither is it just 33%

[russellhq]

True.

I do most of my imaging at ISO 3200 but recent noise tests have shown that with cooling using ISO 6400 is well on.

Ags did some tests on the 1100D and found the noise at 6400 compared to 1600 is over 50% greater. Forgive my ignorance, but what are you hoping to gain from using these high gain settings?

[Chris]

Did Ags cool his cameras for the test like Gina did for hers?

[russellhq]

Did Ags cool his cameras for the test like Gina did for hers?

The link is here: http://stargazerslounge.com/topic/166291-phototool-2012/

They were not cooled but temperature was stable. I would expect though that the same effect would be observed on a cooled camera.

[GlassWalker]

So you're saying a 15Mp colour cam is only 3.75Mp in resolution - don't think the maths works like that - if so then we've all been conned Surely the camera sensor uses two elements eg the luminance provides the image detail and dynamic range and chrominance adds blurred colour 'wash' which the eye find fully acceptable. Ok the bayered resolution can't be 100% but neither is it just 33%

Fundamentally bayer colour filter sensors do not sample all colours at all points, with the information interpolated to produce the nominal full resolution output. As a worst case scenario, the resolution is 25% of nominal! If you have a strongly red or blue subject, with little in the other channels, you're only significantly stimulating 25% of the detectors. e.g. using Ha filter on a bayer sensor, you're throwing away 75% of the sensor output. Removing that bayer filter would get you back to 100% again.

Unfortunately mono cameras where available (astronomical ones too) seem more expensive than their colour counterparts, so if we can remove the colour filter from a DSLR it would enable another level of lower cost imaging.

[Gina]

Ags did some tests on the 1100D and found the noise at 6400 compared to 1600 is over 50% greater. Forgive my ignorance, but what are you hoping to gain from using these high gain settings?

The noise at ISO 6400 should be four times that at 1600 as the gain is 4x (not just 50%). OTOH the noise is temperature dependent, doubling for every 7C rise in temperature so reducing the temperature by 14C makes the noise at 6400 the same as the noise of an uncooled camera at ISO1600. Edited November 1, 2012 by Gina

[russellhq]

The noise at ISO 6400 should be four times that at 1600 as the gain is 4x (not just 50%). OTOH the noise is temperature dependent, doubling for every 7C rise in temperature so reducing the temperature by 14C makes the noise at 6400 the same as the noise of an uncooled camera at ISO1600.

Gina, the signal is also affected by gain. So in principal you're not gaining anything by changing ISO since the SNR stays the same (but it's not as straight forward as that either). There's a very good discussion on the topic here: http://stargazerslou...1600-v-iso-800/ and the posts by IanL are very informative!

Also, Ags showed that the noise response of the 1100D sensor from ISO 100 to ISO1600 behaves as expected but at ISO6400 it does not and instead of four times more than ISO1600 it's more like 6 or 7 times more (if my math is correct).

[Gina]

Fundamentally bayer colour filter sensors do not sample all colours at all points, with the information interpolated to produce the nominal full resolution output. As a worst case scenario, the resolution is 25% of nominal! If you have a strongly red or blue subject, with little in the other channels, you're only significantly stimulating 25% of the detectors. e.g. using Ha filter on a bayer sensor, you're throwing away 75% of the sensor output. Removing that bayer filter would get you back to 100% again.

Unfortunately mono cameras where available (astronomical ones too) seem more expensive than their colour counterparts, so if we can remove the colour filter from a DSLR it would enable another level of lower cost imaging.

I can confirm that the resolution fo Ha is only a quarter of the stated resolution. The 12Mpx 1100D becomes 3Mpx without the advantage of the larger sensor you would expect at that resolution and sensor size. Astro imaging is very different from terrestrial photography. With the latter the lower colour resolution of the human eye compared with luminance can be exploited to make what's really a 3Mpx from the colour point of view appear higher definition.

In other words - YES. The specified resolution is the actual photocells for the whole sensor - and that includes an area round the edge that isn't used for the picture as well.

[Gina]

Gina, the signal is also affected by gain. So in principal you're not gaining anything by changing ISO since the SNR stays the same (but it's not as straight forward as that either). There's a very good discussion on the topic here: http://stargazerslou...1600-v-iso-800/ and the posts by IanL are very informative!

Also, Ags showed that the noise response of the 1100D sensor from ISO 100 to ISO1600 behaves as expected but at ISO6400 it does not and instead of four times more than ISO1600 it's more like 6 or 7 times more (if my math is correct).

That was what I thought but my measurements have disproved it - at least with the sensor at reduced temperature. See my cooling thread (Cooling to near ambient...) for details. Particularly 10,000 secs at ISO 6400 at a degree or two above freezing. Have to say I was very surprised.

[Chris]

The link is here: http://stargazerslou...phototool-2012/

They were not cooled but temperature was stable. I would expect though that the same effect would be observed on a cooled camera.

Thanks for the link:), its a nice program Ags has written there, I wish I was that good with computers that I could write my own Astro software to analyze vignetting and noise, very cool! I agree with his findings about the noise shooting up when usings ISO's above 1600, although Gina's thread makes for very interesting reading, noise seems to be pretty much negated when the cameras cooled, well it does when you consider the length of her sub exposures!!! if noise is that low at 10,000 sec then 5 minutes subs with dither and darks will show very little noise at ISO6400 so why not use 6400 with the proviso that the cameras cooled or you happen to be out side imaging when the ambient temp is -9C .

[Cath]

So you're saying a 15Mp colour cam is only 3.75Mp in resolution - don't think the maths works like that - if so then we've all been conned Surely the camera sensor uses two elements eg the luminance provides the image detail and dynamic range and chrominance adds blurred colour 'wash' which the eye find fully acceptable. Ok the bayered resolution can't be 100% but neither is it just 33%

Well, the real resolution is somewhere a bit above 33%. Interpolation is a routine that simulates the missing coloured pixels with what it thinks would be the pixel value IF it were real, which works just fine for human vision, but it's not good if you want actual real photometric data.

Their is no wideband luminance channel coming from a bayer CFA sensor, only red, green and blue. I think some of the Kodak sensors use a different CFA layout, such as RGBW - which does have a luminance channel, but even that is only 25% of the sensors resolution.

So with a 15 mpix bayer sensor, the raw 15 mpixels (called photosites) has to unfortunately be shared between the red, green and blue channels, some of the pixels give you red intensity, some blue and some green, the gaps are filled in using interpolation to get you back at 15 mpix. So ineffect, 75% of the red pixels, 75% of the blue pixels and 50% of the green pixels (from a bayer CFA sensor) of the RGB image you see are not real, they are guesses of what they might be made by the interpolation routine in the camera.

Wikipedia lists some of the CFA pixel arrangements that are used.

PLUS, the sensing pixels on the sensor are also spaced out (gaps inbetween them where the photons are simply lost), which is where the micro lenses come in - the micro lenses help to focus more of the would-be lost photons directly onto the sensing area. Although their are sensors where the sensing pixel area does cover the entire area, but those sensors cost a lot more, and are not currently generally fitted into consumer cameras.

Edited November 1, 2012 by Cath

[RAC]

You lose alot of dynamic range when using a very high ISO. You will find you have no star colour left as they all just end up bright white. Does not make a nice astro photo in my view. The best DSLR astro shots i've even seen have been done with lower iso's.

[Gina]

You lose alot of dynamic range when using a very high ISO. You will find you have no star colour left as they all just end up bright white. Does not make a nice astro photo in my view. The best DSLR astro shots i've even seen have been done with lower iso's.

Do you? Unfortunately the weather has not allowed me to test the use of ISO 6400. I know 3200 works well with a cooled camera but to be able to pack nearly twice as many subs into the available time would be nice. Only time and weather will tell if this works or whether It's better for me to stick with 3200.

[russellhq]

Yes, but you would only have the same exposure time and surely that's what counts?

Edited November 1, 2012 by russellhq

[russellhq]

I just tried Nitromors on the Playstation Eye sensor. No joy Any other chemicals I could try?

[Kbramley]

You could try Naval Jelly ( phosphoric acid) also known as rust remover - naturally read the precautions on the label before using this stuff.

Keith

[Gina]

Yes, but you would only have the same exposure time and surely that's what counts?

The amount of signal depends on sub exposure time and gain so a 5m sub at ISO 3200 becomes 2.5m at ISO 6400 with the same noise if you reduce the temperature by 7C so in a given total time you can get nearly twice as many subs. (Not quite twice as you have to allow for the sensor reading time between subs.) Hope that clears that up

[allcart]

There is no doubt about it Gina, you will have to run a very controlled test. Same subject, same night. Short subs at 3200 and longer subs at 1600. The processing should reveal the differences if any. If your theory is correct then both pics should look the same with equal processing.