Hunter Hampton
Carbonite Member
Agreed- Which is why the genesis and f35 use oversampled RGB striped sensors to get 1920x1080 with RGB for each pixel.
Sony's answer to the new Panny...
- Thread starter ATL Media Group
- Start date
Agreed- Which is why the genesis and f35 use oversampled RGB striped sensors to get 1920x1080 with RGB for each pixel.
combatentropy
Veteran
Y'all really need to read the link I posted under Graeme's name. Here it is again http://www.reduser.net/forum/showthread.php?t=12026
Barry_Green
Moderator
I'm not a sensor designer -- and I said as much in the AF100 thread.
I was, however, a professional computer programmer, working extensively with graphics, and have more than a fair idea of the processing involved in Bayer demosaic'ing and what it does to the image.
I'm not much in the habit of spouting off on things I have no idea about. Nor do I have much patience for those who do.
So, no -- I'm not a sensor designer. But as a software engineer, and having plenty of discussions with sensor designers, I'm pretty darn confident that I understand very well whats' being discussed here, and I absolutely stand by my assertion that Bayer is not the end-all be-all of image acquisition. It may be the best method for getting images from a single chip, but if you aren't restricted to using only a single chip, you can indeed get better imagery from a 3-chip system.
I really don't understand why this is even a controversy. Use three mysterium-X chips with a prism (after stripping off their bayer filters), and you will get substantially better imagery than you would get from one mysterium-x chip that has a bayer filter on it.
Again, I'm comparing like for like. If you want to go comparing three lesser-pixel chips against one higher-pixel-count chip, well, that's a different argument. But if someone's after the best images that can be made (with no compromises or concessions) they would use three chips -- as every broadcast manufacturer has been doing for decades.
Bayer is only beneficial if you are restricted to a one-chip design.
combatentropy
Veteran
I think that's where we are talking past each other. We both were talking like for like --- just different likes.
If your only constant is lens format (say, 2/3), then three chips will always be better than one.
However, if your constraint is pixel count (say, 6 million), and chip size can vary, then Bayer is the most efficient use of a certain number of pixels. Again, you have to count all the pixels on each of the three chips to compare fairly. That is what Graeme's argument was. He even argues that Sony's RGB stripe method in its big single sensors are not as good as Bayer. Interesting discussion.
Now, practically speaking, the whole discussion is moot. Because in camera design, you never have purely one single constraint.
I don't really understand Graeme's position, because I don't know when you would ever be constrained by total pixel count as he describes.
Yes, one 6 mega-pixel chip is better than three 2 mega-pixel chips, but that's not really a valid comparison, because a 3-CCD camera with 2mp chips is being designed at providing a 2mp final image, while a 6 mp CMOS chip is being designed at provided something closer to a 4 or 5mp final image.
That's like comparing 1 Ferrari to 3 Toyota Camrys. It's not really a valid comparison.
Graeme might have a good point about Bayer being a reasonable and cheaper alternative to 3-CCD, but he is using funny math to illustrate his point.
Last edited:
Yak
Well-known member
I tend to agree with Barry that all things being equal a good 3 chip should produce a superior image compared to a single Bayer chip. But correct me if I am wrong is not one of the advantages of using the micro4/3 Bayer chip that one can use more lenses that would be possible with a prism used in a 3chip camera, this is do to the ability of the micro4/3 chip to be set closer to the lens mount allowing space for different lens adaptor.
Emanuel
Cross-Examiner
Like my country-fellow Mourinho...
Both are 'Special 1'
Jokes aside, here is 1 of his must stickies -- 1 of my favorites:
http://www.dvxuser.com/V6/showthread.php?t=200980
Otis Grapsas
Carbonite Member
I will give you the prospective of a designer although like most of this discussion, it's out of topic.
When you use single sensor you can enjoy the full spectrum from cheap to very expensive cine lenses. You can use sharp SLR lenses. And if your sensor is 2/3" you can use any of the sharp machine vision lenses which are 2x to 4x sharper than many exotic cine lenses in the same aperture. If your pixel size is large such the 5um and larger pixels you get in 2/3" 2megapixel sensor, you get reasonably sharp results with typical SLR lenses and 16mm film lenses which gives you even more options.
If you choose 3xccd in large sensor sizes, the camera cost goes up. 3xccd vs 1xccd at 2/3" could add 6,000 euro to the body cost of a camcorder. In my design that would practically double the retail price. Then you are limited to 3xccd lenses which are expensive and sometimes hard to find. The primes are about 4x to 10x more expensive than the single sensor/film type. The zooms are about 10,000 to 40,000 euro. 3xccd is not without disadvtantages. This automatically turns a 10,000 euro package into a 30,000-40,000 euro package. Very important isn't it? The prism that splits the beams in 3xccd limits the lens maximum aperture. In the ultra expensive 2/3" cameras the limit is f1.4. In the less expensive camcorders (1/3" etc) the limit if f1.6 or f1.8. That's a light loss than one needs to consider because the single sensor camera can use f0.95 or even larger apertures. The 3ccd also has another disadvantage in my book, although many designers disagree. The filters used in 3ccd cameras have very steep edges, very different from those in bayer sensors. Think of a square block (3xccd) vs a rounded hill (bayer). This has an effect in its raw colorimetry and results in very high saturation, especially in red, green and blue. But the best color for me has higher saturation in yellow, cyan, magenta and lower in red, green blue. This is what film looks like and this is easier to achieve with a CCD using bayer filters.
And what do we loose with bayer to get all those advantages? A little resolution depending on the demosaicking algorithm. Distribution only has 4:2:0 chroma so the extra chroma is not that important IMHO. In bayer sensors recording 4:4:4 or RGB or RAW, the 4:2:0 artifacts on common camcorders are avoided. A dslr processed as raw, doesn't look that bad in 1:1 pixel, right? There is technically a little luma loss but 3xccd also has losses due to the prism. We can't get 1080 lines PPH on 1080p formats regardless of the technology used. 3xccd and single ccd both get about 800lines in real world tests, so bayer is not at a luma disadvantage.
The total number of pixels is always limited. The sensor size is given. You want 2/3" for lens compatibility or 24mm wide for 35mm cine lenses compatibility or 35mm for full frame lens compatibility. If you do 4megapixels instead of 2, you get 50% more detail with excellent lenses and perhaps 20% with typical lenses. But your processing cost in acquisition and workflow increases to 2x. Your dynamic range descreases by one stop also. This is probably what Graeme means by the pixel number restriction.
When you use single sensor you can enjoy the full spectrum from cheap to very expensive cine lenses. You can use sharp SLR lenses. And if your sensor is 2/3" you can use any of the sharp machine vision lenses which are 2x to 4x sharper than many exotic cine lenses in the same aperture. If your pixel size is large such the 5um and larger pixels you get in 2/3" 2megapixel sensor, you get reasonably sharp results with typical SLR lenses and 16mm film lenses which gives you even more options.
If you choose 3xccd in large sensor sizes, the camera cost goes up. 3xccd vs 1xccd at 2/3" could add 6,000 euro to the body cost of a camcorder. In my design that would practically double the retail price. Then you are limited to 3xccd lenses which are expensive and sometimes hard to find. The primes are about 4x to 10x more expensive than the single sensor/film type. The zooms are about 10,000 to 40,000 euro. 3xccd is not without disadvtantages. This automatically turns a 10,000 euro package into a 30,000-40,000 euro package. Very important isn't it? The prism that splits the beams in 3xccd limits the lens maximum aperture. In the ultra expensive 2/3" cameras the limit is f1.4. In the less expensive camcorders (1/3" etc) the limit if f1.6 or f1.8. That's a light loss than one needs to consider because the single sensor camera can use f0.95 or even larger apertures. The 3ccd also has another disadvantage in my book, although many designers disagree. The filters used in 3ccd cameras have very steep edges, very different from those in bayer sensors. Think of a square block (3xccd) vs a rounded hill (bayer). This has an effect in its raw colorimetry and results in very high saturation, especially in red, green and blue. But the best color for me has higher saturation in yellow, cyan, magenta and lower in red, green blue. This is what film looks like and this is easier to achieve with a CCD using bayer filters.
And what do we loose with bayer to get all those advantages? A little resolution depending on the demosaicking algorithm. Distribution only has 4:2:0 chroma so the extra chroma is not that important IMHO. In bayer sensors recording 4:4:4 or RGB or RAW, the 4:2:0 artifacts on common camcorders are avoided. A dslr processed as raw, doesn't look that bad in 1:1 pixel, right? There is technically a little luma loss but 3xccd also has losses due to the prism. We can't get 1080 lines PPH on 1080p formats regardless of the technology used. 3xccd and single ccd both get about 800lines in real world tests, so bayer is not at a luma disadvantage.
The total number of pixels is always limited. The sensor size is given. You want 2/3" for lens compatibility or 24mm wide for 35mm cine lenses compatibility or 35mm for full frame lens compatibility. If you do 4megapixels instead of 2, you get 50% more detail with excellent lenses and perhaps 20% with typical lenses. But your processing cost in acquisition and workflow increases to 2x. Your dynamic range descreases by one stop also. This is probably what Graeme means by the pixel number restriction.
Barry_Green
Moderator
You're not wrong -- this is EXACTLY the benefit of using a single-chip design. And if you're using a single chip, you must use something like a Bayer pattern, because sensors are monochrome. You could also use RGB Stripe, or Foveon, but all three systems have their various drawbacks and, when compared side by side, I tend to think Bayer is the best of the single-chip designs.
I'm just saying that as an overall quest for quality, Bayer is nowhere near the top. Its only redeeming value is that if you were constrained to one chip, Bayer's a pretty good way to go on a single chip.
Otis Grapsas
Carbonite Member
If you set a minumum DOF requirement, you will see that most 2/3" shoulder mount cameras are still more capable, because they simply do not need high ISO with the apertures they can practically use. At the same DOF, a 2/3" can use ISO400 instead of ISO 2500 on super35 (f1.4 ISO400 vs f3.5 ISO2500 for same DOF).
The lenses also need to be considered. A 2/3" can get the 125mm super35 angle of view using a 50mm lens. The 50mm is f1.4 but the affordable 125mm is probably f2.8 on an zoom. So it's ISO400 f1.4 vs ISO1600 f2.8.
The larger sensors need higher sensitivity in order to compete because of the limiting shallow DOF and the pricing/availability of fast lenses on larger formats.
Then, there is the global shutter advantage on CCD.
Robert Ruffo
Well-known member
I honesty doubt they would release a dedicated video camera that could not be used to shoot bricks in focus, as is the case with the DSLRs. that would make them a laughing stock.
Dino
Veteran
Otis,
Yes, good info on 2/3" DOF vs. S35 sensitivity requirements. I've owned 17 high-end 2/3" cameras over the past 30 years(12 CCD's), so you could call me a fan!
Panavision and Sony went for CCD for their top-end S35 cameras, but not sure they will remain competitive with the newest CMOS Bayer sensors as far as sensitivity, latitude, resolution and SN ratio. Global shutter artifacts should be less severe in subsequent generations of CMOS sensors as well.
It's interesting that the new Sony 350 CMOS 2/3" camera has very good sensitivity and low noise compared to any 2/3" CCD camera I've seen, 59db, F13. I'm guessing the 59db is done with noise reduction, but I've been impressed by my EX1 in the noise department, even compared to my HPX2700.
Otis Grapsas
Carbonite Member
CMOS can technically lead to better S/N due to its AD conversion approach. CCDs with high pixel rates operate at high frequencies and will produce higher noise vs CMOS. That's another reason the engineers try to stick to 1080p and avoid slow motion when choosing CCD.
Engineers have a different approach when designing cameras in different classes. The EX1 users will probably do lots of low light. The 2700 users or the F35 users will probably spend a lot of time in lighting, so the cameras are tuned for better dynamic range and color in good light. Many would call an F35 noisy vs an EX1. It is, actually. It measures worse in a test using an algorithm to measure. But I would take the texture detail and the glorious saturation over noise performance. I know the methods they use to achieve this S/N on pixels that are that small.
Also, one of they keys of the filmic saturation and rich skin tones is the right film emulation curve. The best types that approach or even surpass film, all produce high noise because they bring the floor up and get higher mid level contrast. Lifeless and noise free or organic and noisy?
Cameras that attempt to use these methods are practically designed for use at 0dB gain.
powervideo
Well-known member
I just bought a Varicam-H which will be here on Monday. I also bought a Nanoflash. I agree which your comments about colour and texture Otis. I've been shooting with the HPX2700 quite a bit and the 720 25P/50P thing is the sweet spot me me at the moment. That said, I would ike an EX1R as a B-camera.
There is a cardinal rule to remember in the world of technology: if they are HONEST and they say it will be out in 6 months, it means that they have begun to develop the technical aspects of it. If they say it will be out in a year, it means they haven't even begun working on it yet. And this is when they are being honest! So in my opinion Sony felt the need to respond at NAB to the revolution happening in the video industry, but this is still only musings in the minds of a few key players in their video division.
hawaj
Well-known member
yep :grin: