4:4:4 10bit single CMOS HD project

[Obin Olson]

I cant save a pre-bayer file but I could send you a 16bit tiff file that has not been touched like the face shot inside

dv3productions.com/test_images/frame2.tif

that is with a cheap cctv lens at 25mm I think better glass will have an effect on the image quality

http://www.dv3productions.com/test_i...udio_test1.jpg
http://www.dv3productions.com/test_i...udio_test2.jpg

shot in our cyc wall with one hard light same image with some "filmLOOK" on the 2nd one

[Steve Nordhauser]

Obin, to get raw data (no Bayer algorithm), do a PIXCI Open/close and close the window. Then do PIXCI Open/close, selct camera format and model and choose the monochrome version. You will get one value per pixel. You can tell - when you zoom it way up it looks like a screen door.

[Les Dit]

Obin,
It would be very interesting to take two pics of the same exact setting, with no motion. Defocussing the lens just a bit. Can you post a tiff sample of that?
This is one of the 'acid' tests for camera noise.
My biggest concern is for noise in lower light settings.
The jpg's you posted look very promising, good work !
-Les

[Jason Rodriguez]

Obin, how far can you underexpose those images to keep the highlights down without inducing too much noise into the image?

Just wondering, especially about the boy with the door in the background. Very nice images, but the highlights seem to clip very harshly, and I'm wondering again if underexposing the image by quite a bit and then lifting it back up in post will still get you a suitable image without too much noise.

BTW, also if these are pure linear files, and you're trying to make them "bright" on the screen so that you can "see" them nicely, then you're not utilizing the full bit-depth of the camera. 10-bits packed into a 16-bit image with padded zero's should look VERY dark, much darker than what you have right now. By making it "brighter", you're sacrificing the room you have in the shadows to increase the brightness of the image, but you end up clipping off when you hit the saturation point of the camera's CMOS. So again, I would think the images should look even darker, especially if they're 10-bit's linearly packed into 16-bit image.

I'm also curious to know how hard you're finding it to capture extensive lengths of video at a time, like a minute or two.

And lastly, the bayer algorithm you're using needs some major improvement, it's totally falling flat on those car shots with the powerlines and the edges on the car-not good. The other images seem to have come out pretty good, but the car shot is horrible. So if anybody is planning on using these cameras for serious work, they're going to have to get a better Bayer demosaic algorithm going.

BTW you might want to check out http://www.dmmd.net/products/products.htm and their Pictura software. Obin, since you have access to a RAW bayer image, you might want to pass it through his bayer demosaic algorithm, like the car shot, and see what it does, if it cleans up the mess of the edges.

Keep us posted, besides the bayer demosaic problems, things are looking pretty good!

[Obin Olson]

guys take a look at this, first image VW-BEETLE is 1300 single cmos camera I shot. 2nd image, varicam is from the Panasonic HD varicam camera.

www.dv3productions.com/test_images/vw_beetle.jpg

www.dv3productions.com/test_images/varicam.jpg


is the varicam worth $98,000 more then the 1300 camera?

(oh ya I shot with a $20 cctv/spy camera lens) ;)

[Jason Rodriguez]

Hey Obin, I'm posting a CC'd version of the TIFF from "Frame2.tif" I was trying to work around the blow-out in the window, trying to make it as smooth as possible. It was quite harsh IMHO. Overall I think it's a very nice "film-ish" look, and yes, I'm quite amazed at the latitude that you can get out of your TIFF file.

I'm still wondering about my previous post though, in how far you can underexpose to get the maximum amount of dynamic range. Do you have a light-meter available to see how many stops you're getting or can get while having a very acceptable image?

http://www.geocities.com/turbochrg/Frame2_CC.jpg

[Obin Olson]

you can underexpose a huge amount, the guy outside,(thats me) was shot with the face almost totaly black and cc back to what it is now

Rob and Rob, take a look at this:

http://www.dmmd.net/products/algorithms.htm

[Steve Nordhauser]

Keep in mind that these are probably captured as 8 bit images since they are 24 bit tiff. That only gives you 256 levels from black to saturated on a pixel. When Obin gets to 10 bit he will be at 1024 levels.

Great shots Obin!

[Steve Nordhauser]

Rob and Rob,
Leadtools has some pretty great tools. As far as I remember, there are no runtime fees so it might be possible to encapsulate their library functions in your own library to make a mostly GPL application.

http://leadtools.com/SDK/Multimedia/...Products-n.htm

I bought one of their general image processing libraries once and it came with a sample program that was like a mini-photoshop. Very impressive.

When someone has the time to really commit to this, let me know.

[James Ball]

Obin,

The viewfinder in the K3 is a reflex type and the mirrored back surface of the shutter reflects the image onto the light path.

I'd suggest:

a) use the beam splitter, 50/50 should work fine. You'll loose one stop and the image through the viewfinder will be 1/2 as bright. A 25/75 will cost you only 1/2 stop but the image will only be 1/4 as bright as looking through the camera.

b) easier, take a video feed and attach an LCD screen.

[Jason Rodriguez]

Took another look at that image that you talked about (the outdoor shot), and yes, I can see the noise which looks as though the shot's been pushed a bit.

If you don't mind me asking, in the pre-color-corrected version, was the background behind you still blown-out? And were these shots at the full ten bits?

If it was still blown out, even with your face in the blacks, then I'm wondering how many stops of dynamic range you're getting. It's probably seven, maybe eight tops if this is the case, which isn't too bad, but not quite as good as I was expecting, since the Viper with a S/N ratio of 54db is at eight to eight-and-a-half good stops of useable dynamic range.

I guess that's pretty reasonable amount though, considering that the Micron chip is actually a consumer digital camera chip, and those little digital cameras only get around seven-to-eight stops max, rather than the ten to eleven that something like the 10D can get.

[Richard Mellor]

thanks for the clips obin . this is looking realy good

[Obin Olson]

Jason the ONLY thing that was blown out was a tiny bit on the truck everything else was underexposed looking from the raw file and the face was in the blacks. I have the 10D and I would say this thing is close to that.....MUCH better then a normal consumer or even professional video camera, all because of 10bit...now I am thinking about using c-mount lenses in the k3 film camera that way I get the FOV that the lens is made for..I could just remove the k3 lens and have a little c-mount sticking out the front...this may work ...I may post some pics of the k3 and ask the group what you guys think

[Jason Rodriguez]

Sorry Obin,

I was just giving my analysis after looking at the file in Photoshop. The whole left side of the image with what appears to be a street and the entire side of the truck is at 255 RGB, which seems clipped to me. The fence is fine, but that's just what I'm seeing after running the eyedropper around the scene in Photoshop.

BTW, if it's not too hard, do you think you could post the RAW file of that outdoor scene with you and the truck? I'd like to see what I can do with it in Combustion if there was as much information there as you're saying.

Thanks.

BTW, the Canon 10D and D60 work off of 12-bit A/D's and have a S/N ratio that's near 70db, so they're much better quality than the Micron chip, actually their specs (but not frame-rate) is very similar to the Altasens 3560, so you may see similar performance as the 10D (which is around 9-10 stops easily) from the Altasens chip. I'm not nocking the Micron, but from the clipping that I'm seeing in your current images, it looks like around 7-8 stops max, which isn't bad, I mean that's where the Cinealta, Viper, etc. are (8-9 stops). The Varicam can supposedly do 11 stops, but I've found that this is pushing the camera to the absolute maximum, and you simply can't safely underexpose that camera without some serious problems with compression artifacts in the shadows when you try to raise them. So even with the Varicam, I think you're getting around 9-stops effective.

Now, the only thing in this estimate is where on the 10-bit scale you're shooting. Are you pushing into the highlights, maximizing the S/N ratio, or are you shooting more into the shadow regions of that range? It seems from the looks of things that you're doing the former, but then I'm not sure how noisey that camera's going to get in the shadows (according to Micron there's only a 45db S/N ratio, which isn't as good as the 54db that the HD cameras have, or the 64-70db that the high-end DSLR's have). I'm not sure if Steve and SI have done some things differently to push that S/N ratio up, but regardless of that, the Micron chip simply isn't what the Canon DSLR chips are, and you can't expect the same performance-but nevertheless you are getting great performance, and having acess to the RAW files probably has a lot to do with that.

BTW, do you know what the approximate ISO of this camera is? Or what is the f-stop rating at 2000lux?

[Rob Lohman]

That are some great looking shots Obin! Obin, Rob & Steve: I've
just returned from my little "holiday" and will be looking at all the
camera "stuff" today... some reading up to do (again).

Obin: how do you have the chip set up now? What lens is in front
and how is it connected to your PC? Can you post a picture of the
"camera" with everything connected etc.?