RED real 4k?

[cinemano]

Hello

I read in a thread (hey it rhymes) that the Red isn't really 4k.. I quote:

"it's a Beyer pattern sensor that delivers a true resolved image of about 2.8 to 3k. It's a 12mp sensor that could potentially deliver 4k in black and white, but when demosaic'd to a color image it's delivering a real-world 3K"

Is that true? what are your thoughts?

[Jeff Kilgroe]

It's 4K in that when shooting in 4K modes, the sensor array is 4096 photosites wide. In a Bayer-pattern sensor, you have one row that alternates photosites as G, R, G, R and so on. On the next row, the photosites alternate as B, G, B, G and on and on.

But it's more complex than just that. While each photosite has a dominant color assign, they still retain some chroma information for other color channels. We also don't know any specifics about the "Mysterium" sensor or entirely how it works. Graeme or Rob could definitely explain this a lot better than I ever could.

As for resolution delivered, there's lots of variables. Glass being one. Any good 35mm lens will resolve more information in terms of lines/mm than needed for the 4K sensor, but other properties of the lens such as light transmission characteristics and contrast or color transmission, etc.. play a role. There are also the de-Bayer algorithms used for processing the data and there's a delicate balance between delivering a sharp image and an image without any aliasing, moire or ringing. I've seen various res chart tests from RED and various results from each. I believe fxphd did one that concluded the RED was about 3.8K nominal res on a monochrome res chart in 4K using a Zeiss super-speed prime (50mm?). Using the RED 18-50mm lens set at 50mm, it showed a bit less contrast and a bit less perceivable resolution, but still upwards of 3.6~3.7K.

No digital camera out there is going to resolve its full imaging resolution without delivering over-sharp, highly aliased or highly pixelated images. At least not as far as I'm aware.

[BASSAM MSSALATIE]

to make a long story short that mean final output is something around 3.8 k:excl:
okay:sick: is ImagerXE telecine machin which scannes film frame at 4k 12 M pixel
quality output the same :waaa: less than 4 k image

[Andrew]

Theoretically you can have 4096 unique horizontal reading and 2048 vertical ones from Bayer CMOS sensor.
Here is some reading, one of many how to do it.
http://www.soe.ucsc.edu/~milanfar/TI...cing_Final.pdf

But the best way to save on heat dissipation and batteries is to oversample the 4K by factor of 3 so 4096X2048 = 8MP
Having 24 MP sensor will make it very smooooth…..
Also it will make it possible to get 4K RGB signal right out of the sensor logic without going in to the fast and hot circuits. The other question is the dynamic range, more pixels at the same size of the sensor smaller pixels and less light.

So we have to wait for progress in this technology for 24MP

Canon Ds Mark III has 21MP but the sensor size is double the size of Misterium.
Though I wouldn’t mind to use full size sensor with Canon L full size sensor lenses.
Then you can forget about all these cine type lenses, these lenses will not work with full size sensor.

[Chris Kenny]

Remember there's also an optical low-pass filter in there to prevent aliasing. Nyquist says (or so my understanding goes) that to get "real" 4K with no aliasing, you'd need at least an 8K sensor.

Once you get into resolutions like what Red delivers with its 4K bayer sensor (which is, remember, well above what you get from a 35mm release print), counting every pixel doesn't matter all that much. The question becomes, does it look good on a large screen? I haven't heard anyone complaining on that score.

[Brook Willard]

There are OLPFs on every sensor... RED or not. That said, they're all built to different specifications. An OLPF from a 20D would fail miserably on the RED [in motion], as an example.

[Andrew]

When you have 3:1 oversampling you don't need much of OLPF filtering.

[Jannard]

This topic has been covered many times before... please do a quick search.

The RED ONE sensor captures 4096 pixels so it is a 4K camera. Because of the Bayer methodology, final resolution is greater than 3K. A 4K film scan does the same thing. It captures 4K pixels, but the real resolution is maximum of 3.2K on slow film (as measured by one of the top studios). The difference is that most film is not shot ISO 50. And the RED image is much cleaner than an equivalent ISO 320 (or 1000) film stock.

Film has more latitude (for now). It also has the costs of film, processing and transfer associated with it.

We have always said that RED is an alternative to film. We think a very good alternative. Resolution has never been an area of concern in comparison to film.

Jim

[Graeme Nattress]

8k would be needed to get 4k rez on all chroma, but would give at least 6k on luma. However, you'd need lenses with a very high MTF, and you'd also (all other things being equal) loose dynamic range unless the sensor was made proportionally larger.

They key things to note are;

You cannot get "x" resolution from a sensor with "x" pixels unless you allow for aliasing. Aliasing is bad, especially on moving images, so we optically low pass filter to avoid this, giving us maybe 80% or so of "x" as a safe maximum. This would apply to any camera of any technology. There is another loss due to the Bayer pattern, but you should on any system get around > 70% of "x" with both Bayer and OLPF present. On some of my test charts I've read up to 78%, but obviously the very highest resolutions have a very low modulation due to lens MTF and OLPF. However, the images do look high resolution when projected, noticeably higher than HD or even the best release print. They also look better than 4k scaled to 2k and back up to 4k (or native 2k scaled up), thus really validating the approach.

Just to add into what Jim is saying, the proof is in the pudding, and on the big screen our pudding looks and tastes great.

Graeme

[Andrew]

Originally Posted by Chris Kenny

Remember there's also an optical low-pass filter in there to prevent aliasing. Nyquist says (or so my understanding goes) that to get "real" 4K with no aliasing, you'd need at least an 8K sensor.

Once you get into resolutions like what Red delivers with its 4K bayer sensor (which is, remember, well above what you get from a 35mm release print), counting every pixel doesn't matter all that much. The question becomes, does it look good on a large screen? I haven't heard anyone complaining on that score.

Try these picture samples at 21MP

http://web.canon.jp/imaging/eosd/eos..._sample-e.html