4:4:4 10bit single CMOS HD project

Ben Syverson · July 29th, 2004, 11:15 AM

@Les: I get a kick out of terms like pedestal and IRE black, when they are applied to computer images! They even use "Mhz" to talk about resolution sometimes!

Pedestal and IRE black, along with IRE white values, are totally and completely irrelevant with these cameras. It is much more like exposing a negative.

The "pedestal" is controlled by an "offset" control (in my case, two offset controls -- one for each G channel). You generally set your main offset once, to the black level of your liking, and then adjust the other offset every once in a while to make sure it's matching up with the first offset.

It's relatively easy to overexpose if you try to make the on-screen image look "normal." Like Jason has mentioned many times, the raw image should look pretty dark. Only specular reflections and bright highlights should be approaching the max value (be it 255 or 1024 or 4096). I adjust my camera so the brightest thing in the scene is just under 255, so I can be sure that I'm not clipping them at all.

In post (or on the camera hardware, at least partially), you apply your gamma and move your white points down to increase exposure. This is where 8bit starts to fall apart. You really need at least 10 bit to make these corrections gracefully. After that, you increase the chroma/saturation however you want to do that, and then you should have a fairly "normal" looking image. This is pretty much what consumer cameras do internally.

Mhz has been used to talk about resolution on some HD cameras, but I think its a poor benchmark. The idea is, the more pixels, the faster the sensor/camera's clock has to run to read them all out. So rather than say 2.1 million pixels, you say XXmhz. The problem is that no one but engineers care what clock rate the camera is running at... People just want to know what the final resolution and frame rate is. :)

- ben

Steve Nordhauser · July 29th, 2004, 11:22 AM

Rob: I would be very interested to know what problems there are with standard lenses with a 3 chip. The only issue that the prism manufacturer discussed with me was the length of the optical path vs the back focal distance.

Juan:
Can't do that right now - I have to work on the 1920HD to get it out. The Jackson shot was taken with very good lighting and gain and offset correction.

Obin:
I guess if you can saturate the sensor without smears, you can get better dynamic range by shooting the full range but they are both 10 bit cameras.

Rob Scott · July 29th, 2004, 11:40 AM

Quote:

Steve Nordhauser wrote:
I would be very interested to know what problems there are with standard lenses with a 3 chip.

Well, I would trust the prism manufacturer over myself! What I had heard was that the prisms often had odd optical behaviors -- red fringing, for example, that required additional optics in the lens to compensate. As I recall, it was about the Zeiss lenses made specially for the Thompson Viper -- they cost $115K or so. I can't find the reference now, however, so I might be totally blowing smoke (I hope so).

Juan M. M. Fiebelkorn · July 29th, 2004, 11:54 AM

Well, I undertand that we are not working with normal video, but for film the IRE standard is also used.TEC from KODAK gives its values in IRE scale so I stick with that.
When you go from digital to film you use TEC, too.
Anyway if it isn't correct it works very well.... :)

Wayne Morellini · July 29th, 2004, 12:03 PM

So that is why they cost so much (as well), puts a whole new spin on keeping single chip.

Jason Rodriguez · July 29th, 2004, 01:00 PM

4-5 stops is like a bad video camera

You can see in the images I posted what 9-10 stops looks like (Viper).

Obin Olson · July 29th, 2004, 02:15 PM

Jason I don't think I am getting 4-5 stops...how on earth could I shoot that windos shot and have the inside room with NO lights and the outside in the range it is??

Juan M. M. Fiebelkorn · July 29th, 2004, 02:56 PM

I don´t know what time of the day it was, but that shot won´t give better results on 35 mm...

I still think it isn´t correctly exposed (nothing personal Obin, just my point of view).I've been playing with the 16 bit tiff in Combustion with no success, shadows too deep and highlights too high....:(

Jason Rodriguez · July 29th, 2004, 03:19 PM

That's what was in those MacBeth charts-about 5 stops.

You could do some comprehensive tests yourself, check out the shot of the different charts with the Viper.

In that shot, they set up three charts with three levels of light in between with a light meter setting for each. The one on the left was lit 3 1/2 stops lower (according to 18% grey), the middle was at EV 8, and the one on the right was a little over a stop hotter ( according to 18% grey again). That would put the white chip on the right hand side one stop hotter than the center white chip, and the black chip on the left hand side is 3 1/2 stops lower than the black chip in the middle. Because you have a 4 1/2 stop lighting difference between the dark and the light, and the black chip is another 2 stops below the EV 4.6 of the dark side, that gives you 6 1/2 stops. On the highlight side you have another 3 stops above the middle grey chip, for a total of 9 1/2 stops in that scene. That's quite a bit of dynamic range. You can try that test too and see what you get.

That should give you a pretty objective measure of what the chips can and can't do, instead of thinking you're getting a lot by shooting hot windows.

Rai Orz · July 29th, 2004, 03:51 PM

Steve, is there a camera with global shutter available?

Steve Nordhauser · July 29th, 2004, 06:27 PM

Rai:
At the resolution/frame rates that are being used in cinematography, we only have IBIS-5 cameras - the SI-1280F in USB 2.0, camera link and gigabit ethernet. If Ben gets everyone revved up on IBIS-5, we have a 12 bit version using an external A/D for faster rates and lower noise.

The only other global shutters we have are a VGA at 250fps and a 2Kx2K large pixel, 15fps (2Kx1K@30fps) MONO ONLY. Gosh darn. Something like $1M to get a color mask done and fabbed - the sensor company has no plans to do it. It would make a bodacious 3 chip camera though...... We have to see what the image quality looks like. It is in layout now.

Juan M. M. Fiebelkorn · July 29th, 2004, 08:33 PM

Steve,
Could you tell me what's the usual way you use to remove the FPN from a sensor?

Ben Syverson · July 29th, 2004, 09:43 PM

Juan, what does a 2 chip camera get you? Would that be 2 bayer-filter sensors, each recording at a different exposure setting in order to form 1 HDR image? Clearly there's no way to get color from 2 b&w chips, so you must be talking about HDR from 2 color sensors...

Obin Olson · July 29th, 2004, 10:25 PM

Jason your right! and untill we get our software running IN COLOR I will wait to test...

I have an update today that has the software PREVIEWING in COLOR! awesome...but the framerates are very low when I hit record...so I will give my coder a bit more time to get that up ;)

Jason Keenan · July 29th, 2004, 10:29 PM

Hey there,

Just thought I'd put something out there. Just wondering if there was a 'cheap' compatible camera available that could capture at say, PAL resolution (or slightly larger). When I say cheap I mean at least 1/4 of the cost of the HD camera you are looking at here. There is a 640x480 camera on the site but due to the high framerate it gets, I'm assuming it's still expensive.

I'm pre-empting the modularity of the 'Obscura' system. The idea that you could cheaply use the same software etc. to capture video comparable to higher end prosumer cameras (eg XL1) then be able to upgrade to the HD camera with little pain.

The idea of being able to get a progressive scan system with the ability to swap lenses for $1000+ US, plus the abilty to upgrade to HD for the price of a new XL1 would be attractive I think.

Is there such a camera and would the stuff learned here translate over to the said camera.

Raavin (Use my handle to save confusion with the other Jason)

July 29th, 2004, 11:22 AM	#1217
Steve Nordhauser Silicon Imaging, Inc. Join Date: May 2004 Location: Troy, NY USA Posts: 325	Rob: I would be very interested to know what problems there are with standard lenses with a 3 chip. The only issue that the prism manufacturer discussed with me was the length of the optical path vs the back focal distance. Juan: Can't do that right now - I have to work on the 1920HD to get it out. The Jackson shot was taken with very good lighting and gain and offset correction. Obin: I guess if you can saturate the sensor without smears, you can get better dynamic range by shooting the full range but they are both 10 bit cameras. __________________ Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com

July 29th, 2004, 06:27 PM	#1226
Steve Nordhauser Silicon Imaging, Inc. Join Date: May 2004 Location: Troy, NY USA Posts: 325	Rai: At the resolution/frame rates that are being used in cinematography, we only have IBIS-5 cameras - the SI-1280F in USB 2.0, camera link and gigabit ethernet. If Ben gets everyone revved up on IBIS-5, we have a 12 bit version using an external A/D for faster rates and lower noise. The only other global shutters we have are a VGA at 250fps and a 2Kx2K large pixel, 15fps (2Kx1K@30fps) MONO ONLY. Gosh darn. Something like $1M to get a color mask done and fabbed - the sensor company has no plans to do it. It would make a bodacious 3 chip camera though...... We have to see what the image quality looks like. It is in layout now. __________________ Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com

July 29th, 2004, 11:15 AM	#1216
Ben Syverson Major Player Join Date: Mar 2004 Location: chicago Posts: 434	@Les: I get a kick out of terms like pedestal and IRE black, when they are applied to computer images! They even use "Mhz" to talk about resolution sometimes! Pedestal and IRE black, along with IRE white values, are totally and completely irrelevant with these cameras. It is much more like exposing a negative. The "pedestal" is controlled by an "offset" control (in my case, two offset controls -- one for each G channel). You generally set your main offset once, to the black level of your liking, and then adjust the other offset every once in a while to make sure it's matching up with the first offset. It's relatively easy to overexpose if you try to make the on-screen image look "normal." Like Jason has mentioned many times, the raw image should look pretty dark. Only specular reflections and bright highlights should be approaching the max value (be it 255 or 1024 or 4096). I adjust my camera so the brightest thing in the scene is just under 255, so I can be sure that I'm not clipping them at all. In post (or on the camera hardware, at least partially), you apply your gamma and move your white points down to increase exposure. This is where 8bit starts to fall apart. You really need at least 10 bit to make these corrections gracefully. After that, you increase the chroma/saturation however you want to do that, and then you should have a fairly "normal" looking image. This is pretty much what consumer cameras do internally. Mhz has been used to talk about resolution on some HD cameras, but I think its a poor benchmark. The idea is, the more pixels, the faster the sensor/camera's clock has to run to read them all out. So rather than say 2.1 million pixels, you say XXmhz. The problem is that no one but engineers care what clock rate the camera is running at... People just want to know what the final resolution and frame rate is. :) - ben

July 29th, 2004, 11:54 AM	#1219
Juan M. M. Fiebelkorn Major Player Join Date: Jun 2004 Location: Buenos Aires , Argentina Posts: 444	Well, I undertand that we are not working with normal video, but for film the IRE standard is also used.TEC from KODAK gives its values in IRE scale so I stick with that. When you go from digital to film you use TEC, too. Anyway if it isn't correct it works very well.... :)

July 29th, 2004, 12:03 PM	#1220
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	So that is why they cost so much (as well), puts a whole new spin on keeping single chip.

July 29th, 2004, 01:00 PM	#1221
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	4-5 stops is like a bad video camera You can see in the images I posted what 9-10 stops looks like (Viper).

July 29th, 2004, 02:15 PM	#1222
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	Jason I don't think I am getting 4-5 stops...how on earth could I shoot that windos shot and have the inside room with NO lights and the outside in the range it is??

July 29th, 2004, 02:56 PM	#1223
Juan M. M. Fiebelkorn Major Player Join Date: Jun 2004 Location: Buenos Aires , Argentina Posts: 444	I don´t know what time of the day it was, but that shot won´t give better results on 35 mm... I still think it isn´t correctly exposed (nothing personal Obin, just my point of view).I've been playing with the 16 bit tiff in Combustion with no success, shadows too deep and highlights too high....:(

July 29th, 2004, 03:19 PM	#1224
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	That's what was in those MacBeth charts-about 5 stops. You could do some comprehensive tests yourself, check out the shot of the different charts with the Viper. In that shot, they set up three charts with three levels of light in between with a light meter setting for each. The one on the left was lit 3 1/2 stops lower (according to 18% grey), the middle was at EV 8, and the one on the right was a little over a stop hotter ( according to 18% grey again). That would put the white chip on the right hand side one stop hotter than the center white chip, and the black chip on the left hand side is 3 1/2 stops lower than the black chip in the middle. Because you have a 4 1/2 stop lighting difference between the dark and the light, and the black chip is another 2 stops below the EV 4.6 of the dark side, that gives you 6 1/2 stops. On the highlight side you have another 3 stops above the middle grey chip, for a total of 9 1/2 stops in that scene. That's quite a bit of dynamic range. You can try that test too and see what you get. That should give you a pretty objective measure of what the chips can and can't do, instead of thinking you're getting a lot by shooting hot windows.

July 29th, 2004, 03:51 PM	#1225
Rai Orz Regular Crew Join Date: Jun 2004 Location: Germany Posts: 136	Steve, is there a camera with global shutter available?

July 29th, 2004, 08:33 PM	#1227
Juan M. M. Fiebelkorn Major Player Join Date: Jun 2004 Location: Buenos Aires , Argentina Posts: 444	Steve, Could you tell me what's the usual way you use to remove the FPN from a sensor?

July 29th, 2004, 09:43 PM	#1228
Ben Syverson Major Player Join Date: Mar 2004 Location: chicago Posts: 434	Juan, what does a 2 chip camera get you? Would that be 2 bayer-filter sensors, each recording at a different exposure setting in order to form 1 HDR image? Clearly there's no way to get color from 2 b&w chips, so you must be talking about HDR from 2 color sensors...

July 29th, 2004, 10:25 PM	#1229
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	Jason your right! and untill we get our software running IN COLOR I will wait to test... I have an update today that has the software PREVIEWING in COLOR! awesome...but the framerates are very low when I hit record...so I will give my coder a bit more time to get that up ;)

July 29th, 2004, 10:29 PM	#1230
Jason Keenan New Boot Join Date: Jul 2004 Posts: 19	Hey there, Just thought I'd put something out there. Just wondering if there was a 'cheap' compatible camera available that could capture at say, PAL resolution (or slightly larger). When I say cheap I mean at least 1/4 of the cost of the HD camera you are looking at here. There is a 640x480 camera on the site but due to the high framerate it gets, I'm assuming it's still expensive. I'm pre-empting the modularity of the 'Obscura' system. The idea that you could cheaply use the same software etc. to capture video comparable to higher end prosumer cameras (eg XL1) then be able to upgrade to the HD camera with little pain. The idea of being able to get a progressive scan system with the ability to swap lenses for $1000+ US, plus the abilty to upgrade to HD for the price of a new XL1 would be attractive I think. Is there such a camera and would the stuff learned here translate over to the said camera. Raavin (Use my handle to save confusion with the other Jason)