Debayering a DSLR's Bayer matrix.

[12dstring]

Here's a video showing how the CFA scrapes away.

I used the metal pick which, although can scrape the layer below, is a lot finer so it's easier to see what's happening (and I was only using the broken 'practice' sensor).

[billhinge]

great stuff, what magnification was that ?

how come there appears to be 3 colour areas eg at 33s on video

on the left there is a purple colour (micro lens?), the green colour in centre (CFA?) and a dark colour top right (trick of the light or glass?)

[12dstring]

I've added annotations, but in case they're not working:

4x magnification at the beginning, switching to 10x at 1:23

The section on the left at the beginning is untouched, with microlenses. There's a small area just left of centre at the bottom where the microlenses are 'squashed' but not removed.

The dark section in the top right is with the microlenses totally removed but the CFA still there, and the yellowy/green section in the bottom right is with the CFA already removed.

When it switches to 10x it's just CFA minus microlenses (save for a few in the top left) at the top, and CFA removed at the bottom.

[billhinge]

Just watched it again with your notes in the last post, sorry I missed the annotations first time and it makes sense, very good

[tonedeaf]

very nice video!

have to say that i broke my ps3 eye cmos while trying to debayer

at least i learned something...

for all the ps3 eye protective glas is easy to remove, not like on the eos 1100d

[russellhq]

very nice video!

have to say that i broke my ps3 eye cmos while trying to debayer

at least i learned something...

for all the ps3 eye protective glas is easy to remove, not like on the eos 1100d

I broke 3 PS3 eyes and gave up. In hindsight, I think the stanley knife was the wrong tool for the job...

[tonedeaf]

I broke 3 PS3 eyes and gave up. In hindsight, I think the stanley knife was the wrong tool for the job...

it seems to be that the ps3 eye bayer matrix is constructed in an other way...even after scraping with a "self" built scraping tool nothing changed....

maybe using a microscope like 12dstring is the only way to go

[russellhq]

Tonedeaf, I did see a changed when I tried it, but had to scrape quite hard to get the CFA off. I think that's how I ended up killing the sensor. That's why I think a chemical approach would be safer. Less chance of damaging the sensor.

[tonedeaf]

that was the killing of mine...nothing changed, took a screwdriver (very small one) and scratched lightly ...hurrray a change! used it, made a pause connected the camera -> no image and the sensor´s going hot...

have to try it someday with a 450D, but first i´ll try to by a normal webcam to debayer it, cause i want to try the qhy5II as a dso cam, so i´ll need a "second" guiding cam

[billhinge]

Just picked up a second hand bressar microscope off ebay - can alway sell it back and get a better one if I like microscopy

Plan to have a another couple of tries over summer, felling confident

[pixueto]

And this procedure isn't really damaging the glass surface under the CFA and on top of the diodes? Shouldn't this glass end up full of scratches?

[12dstring]

It depends what the hardness of the scraping tool is.

With a microscope it's easy to see if you've scratched the glass layer below. I started out with a metal tool as it was smaller and removed the CFA very easily but, as you can see in the video above, it does occasionally scratch the glass underneath. But as I only used this on the first broken sensor, I'm not sure how much this degrades the image.

I soon moved on to a sharpened plastic (I think perspex) tool, which has to be resharpened every now and then to be effective, but can't scratch the glass below.

That's why I think a chemical approach would be safer. Less chance of damaging the sensor.

Has anyone has any success with chemically removing a DSLR CFA? It would be great if it worked, I just remain a bit skeptical how effective it would be..

[billhinge]

There was a claim from 2008

see here http://photo.net/casual-conversations-forum/00RmUd?start=30

http://en.wikipedia.org/wiki/Methyl_ethyl_ketone

never tried it myself though

[12dstring]

Thanks, that's interesting. Although earlier in the thread:

I won't be doing any Canons. There are three of us (that I know of) who have attempted monochrome conversions. My Nikon D100 was successful. A friend's Nikon D2X was successful. These were done with failty common aliaphatic solvents. A Canon 20D was reported to have met with dismal failure, either the microlenses or the Bayer filters (or both) are epoxy based, and cannot be dissolved with readily available chemicals.

So perhaps it's not so simple with Canons, or some of them anyway. I'll keep experimenting with my dead sensor so will throw some solvents at it some time.

I've cleaned up my other one a bit more, and couldn't resist a bit more debayering. I was going to announce the bad news that I'd somehow severed another gold wire, but the good news is the sensor still somehow works. Hopefully it's just an extra ground connection and not something too important...

I'm thinking about putting a dab of electronics silicone on the wires to give them some resistance from accidental touch. It's way too easy to brush one of them and then it's game over.

[12dstring]

Proof it still works, and an outside shot. You can view the full image here: www.12dstring.me.uk/images/equipment/350d/debayering/0008-full.jpg

Also attached is a new flat, after a bit more scraping and some cleaning with a fine brush and compressed air. Will finish the right edge and do a proper clean at the weekend hopefully. The rest can be cropped out.

[billhinge]

Did you see any resolution (or other improvements eg ev sensitivity ) with the monochrome camera ?

[12dstring]

At full scale there's definitely a resolution improvement (comparing to the same image with the interpolation algorithm applied). You can see finer detail, however the interpolation has the effect of appearing to reduce noise a bit due to the blurring, so it does look slightly noisier in the details.

Something to note though is that unless you are cropping a small portion of the image, or printing out in a large format you probably won't make use of this as, if like me, you downsize the final image for displaying on the web/computer.

Sensitivity under normal light isn't much changed, at least using the camera's auto metering results in an image that isn't under or over exposed. As mentioned earlier removing the filter array pretty much undoes the sensitivity loss of removing the microlenses with the net effect being not a lot of change in full spectrum light.

That said, my aim with this (apart from just having fun) is to have a better narrowband camera. Next time I can I'll take some shots of the sun in h-alpha with this camera and an unmodded one. This should show off the resolution improvement and big sensitivity improvement for narrowband use.

[billhinge]

Cool and well done

reason I asked about was I was reading this panasonic article about a new filterless sensor which quotes absorption by the CFA being about 50% to 70% (i.e. only 30/50% transmitted) http://www.imaging-resource.com/news/2013/02/05/bye-bye-bayer-panasonic-claims-new-sensor-tech-ends-color-filter-light-loss

[tonedeaf]

huch panasonic is working!

i think the given value of 50-70% might be right...

but on the other side how much loss is there w/o the microlenses 30-40% or more?

there is no info given anywhere...just the photos that where here made...

the comparsion shots will reveal the truth

[russellhq]

Cool and well done

reason I asked about was I was reading this panasonic article about a new filterless sensor which quotes absorption by the CFA being about 50% to 70% (i.e. only 30/50% transmitted) http://www.imaging-r...lter-light-loss

Is that 50-70-% of all light? Or 50-70% of the light the filter's designed to transmit? I.e. does a red filter absorb 50-70% of the red light or 50-70% of the overall light?

I think there's a significant difference.

[12dstring]

It would be each filter absorbs 50-70% of white light and transmits 30-50%. The transmittance of red light through a red filter would be much higher.

For an 'ideal' filter array it would be 33% transmittance of white light in each filter, as for example the red filter would transmit all red light, and no green or blue light. The Baader RGB filters for example are pretty close to that.

However you can see the response from a Canon DSLR isn't quite so simple as there aren't sharp cutoffs, and lots of overlap between filters, and they're not all equal (note that image also factors in the response of the sensor just to confuse matters).

[billhinge]

Is that 50-70-% of all light? Or 50-70% of the light the filter's designed to transmit? I.e. does a red filter absorb 50-70% of the red light or 50-70% of the overall light?

I think there's a significant difference.

I'm guessing but maybe thats why there is a 20% spread?

I see 12dString provided an answer

Edited May 16, 2013 by billhinge

[russellhq]

At full scale there's definitely a resolution improvement (comparing to the same image with the interpolation algorithm applied). You can see finer detail, however the interpolation has the effect of appearing to reduce noise a bit due to the blurring, so it does look slightly noisier in the details.

Something to note though is that unless you are cropping a small portion of the image, or printing out in a large format you probably won't make use of this as, if like me, you downsize the final image for displaying on the web/computer.

Sensitivity under normal light isn't much changed, at least using the camera's auto metering results in an image that isn't under or over exposed. As mentioned earlier removing the filter array pretty much undoes the sensitivity loss of removing the microlenses with the net effect being not a lot of change in full spectrum light.

That said, my aim with this (apart from just having fun) is to have a better narrowband camera. Next time I can I'll take some shots of the sun in h-alpha with this camera and an unmodded one. This should show off the resolution improvement and big sensitivity improvement for narrowband use.

The quality of the bayer matrix image is also dependent on the de-bayer routine used. This article was posted in another topic and makes good reading on different de-bayer techniques. Using the VNG method, the bayer matrix image is pretty much as good as the original.

http://www.stark-lab...ayering_API.pdf

[jdouglas76]

I want to thank everyone that has posted on here about removing the bayer filter. It has helped me quite a bit in my experimenting.

I have been having a difficult time getting my raw images converted with dcraw. They are overexposed and if I adjust levels I get a very grainy image. I have just been using DCRAW -D image.cr2 to convert. I have only so far removed the glass on point and shoot digitals. I have been using a hot air rework station to heat up the glass on the sensors. So far I have removed the glass from five small sensors without to much trouble. The one I converted is a canon s2is and I had to use CHDK add on to get RAW from it. I am wanting to convert a canon g12, canon rebel or nikon d80 next. Someone said that C2H4Cl2 1,2-dichlorethane will remove the sensor bayer filters. I have not yet checked to see where I can get this toxic solvent. I do plan on trying it at some point.

I would love to know what software maxmax is using to produce the nice test images on their web page. I also wonder if CHDK could be modified to produce a good clean image automatically in camera without post processing.

I did not take the picture in ideal lighting but the jpeg of this image still looked far better. This is the best image I've been able to make. I am hoping I can learn of a better raw conversion before experimenting on more expensive cameras.

[12dstring]

Great work! What method did you use for removing the filter?

For dcraw I use "DCRAW -d -4 -T image.cr2" - which outputs a 16 bit linear TIFF file that just needs a little stretching in photoshop.

A heads up for anyone in the UK; 'camnology' on eBay is selling camera parts, including CMOS sensors from Canon DSLRs. They're listed as untested/for spares and repairs, but I picked up a cheap 350D sensor that works fine. Might be worth a shot if you want to have a go with less risk.