anamorphic lenses take 2

[Martin Drew]

Stephen you are right that there is going to be more resolution from three 9.2 megapixel panels than from one 9.2 mega bayer panel, but I am sure you would accept that you can't just multiply the number of pixels by the number of panels to give you a figure for resolution. Even if there is some spatial offset between the panels it still wouldn't give you 3 x the resolution.

And I reckon I could guess at an answer to your second question. The 4K RGB scan will probably have the upper hand. Graeme previously said that a bayer array can resolve at over 70% of the pixel count. So a 4.5K bayer should give you a minimum of 3.15k resolution. That said the 4K scan won't resolve 4K of course because it will require OLPF.

Getting back to the original thrust of the thread though 4.5K Bayer in the Red with a 2:1 anamorphic lens should have sufficient resolution (2540 x 3035) for downsampling to 2K , even allowing for the necessary cropping.

M

 

[Stephen Williams]

Stephen you are right that there is going to be more resolution from three 9.2 megapixel panels than from one 9.2 mega bayer panel, but I am sure you would accept that you can't just multiply the number of pixels by the number of panels to give you a figure for resolution. Even if there is some spatial offset between the panels it still wouldn't give you 3 x the resolution.

And I reckon I could guess at an answer to your second question. The 4K RGB scan will probably have the upper hand. Graeme previously said that a bayer array can resolve at over 70% of the pixel count. So a 4.5K bayer should give you a minimum of 3.15k resolution. That said the 4K scan won't resolve 4K of course because it will require OLPF.

Getting back to the original thrust of the thread though 4.5K Bayer in the Red with a 2:1 anamorphic lens should have sufficient resolution (2540 x 3035) for downsampling to 2K , even allowing for the necessary cropping.

M

Hi Martin,

Basically in agreement with you.

Interested to see how Anamorphic Red or cropped Red looks like. I guess I dont have to wait that long!

Stephen

 

[Joshua Provost]

Why not? There is far mor data from a 9.2 megapixel RGB than from a 9.2 Bayer.

It will be interesting to compare a 4k RGB scan to a 4.5K Bayer sensor and see which has the most resoloution.

Stephen,

You are absolutely correct, all other factors being equal, there is more detail to be had from a 3CCD array than a 1CMOS (or 1CCD) array.

I only mention it because I have seen a lot of people mistakenly assume that you can simply multiply the resolution of each CCD in a 3CCD array and arrive at the resolution. This discounts the fact that the channels are combined to create a single image with the resolution of just a single CCD. I know you know way better than that, but I didn't want to let that slip.

It would be more fair and accurate to say that a 1CMOS has less resolution than its quoted MP spec, than to say that a 3CCD has more resolution than a single individual sensor.

So, in this case the comparison is much closer than 12MP vs. 27MP would indicate. Plus, the RED promises to have a killer debayer algorithm.

Josh

 

[Jeff Brue]

The D20 may downsample (1.5x in 'TV mode' to go from 2880 pixels wide to 1920) but that certainly does not eliminate artefacts. Attached is a crop of a grab from a test I did with a D20 last week, captured 10-bit uncompressed straight off the camera, not via tape. It shows very noticeable Bayer artefacts.

No Bayer sensor is immune to those artefacts, but hopefully with the RED doing a 2:1 downscale to give 1080p, they should be greatly minimised, as each Bayer cluster of four pixels will be scaled to a single 1080p pixel. At least that's my assumption.

I'd just to add on a bit to that. I wouldn't say no bayer sensor I'd say no bayer algorithm. I've been hearing of some fairly ingenious algorithms that use motion picture estimation to actual generate the proper demosaic on a frame by frame basis. Again its a thing that could never been done on camera on a realtime HD-SDI out, but if it means 5 seconds a frame on a modern machine...well its just really cool math instead of a bad sensor.

Also in regards to sub sampled color. I work with the Viper and the S2 workflow at the camera house in Hollywood. I have to say the footage is gorgeous on a day in day out basis. The things that it lets vfx do are great, and because you don't have to deal with things like compression, gate weave, film scratches, whatever gets in the way. New applications for post like optical flow can become incredibly more useful.

Thats part of why I'm terriblly interested in the idea of capturing a wavlet compressed bayer pattern like graeme is doing because all of those interesting algorithms will still be eminently possible. Don't get me wrong I have a feeling we'll still get the optical out ports, but Red on paper is going about this the right way.

Jeff Brue
Technical Director
Digital Film Company