HDR video concept for future 5D Mark III successor

Apefos Adapter

Apefos Adapter

Well-known member
The pixel count of 5D Mark III is very inteligent and smart. For 16x9 video it will be 5760 x 3240 pixels.
5760 = 1920 x 3 and 3240 = 1080 x 3
So probably the camera will take one pixel from each 9 pixels block. This will probably avoid moire/aliasing and increase resolution.

If Canon could develope a 5D Mark IV with Dual Digic 5+ processors it could do HDR video using the single central pixel from 9 pixels block to do one exposure and use the 8 surrounding pixels from the 9 pixels block to do the second exposure. Using more pixels to get the information of a single pixel can generate more electrons than using just one pixel, so this can be used to do HDR video. This way there is no need to use different shutter speeds to get the two different exposures. Also second exposure using the 8 surrounding pixels from the 9 pixels block could use 2 or 4 or 8 pixels so it could do a 1 or 2 or 3 fstops different exposure when comparing to the exposure from the central single pixel. Both images would be combined in a wider dynamic range image.

HDR video straight from camera is the big step improvement filmmakers are waiting for.
 
I'm not sure it works in the way you're saying...

in any case, for me the way forward is to use expanded Bayer patterns: RGBW or RGBWK
a 3x3 basic structure with 2xR 3xG 2xB 1xW 1xK would be nice
a 2x4 basic structure with 1xR 1xG 1xB 4xW 1xK would be even nicer http://www.similaar.com/foto/bayer/rgbwk.html
that can "easily" lead to 3 or 4 stops higher DR than a standard RGB Bayer pattern
of course, 4 stops higher DR on a 4:2:0 8-bit codec doesn't make a lot of sense, so let's save that and RAW video recording for the next generation :)
 
Your idea is great. But why so many white pixels in the 2x4 RGBWK pattern? It could be a 6 pixels patern (2x3), 1r, 1b, 2g, 1w, 1k or 1r, 1b, 1g, 2w, 1k. And i think the pixel block must be square, so in this 6 pixels block each pixel must be rectangular to make the block square, or need to use the 9 pixels block, 3x3 estructure using square pixels to get a square block. Also the block could be 5 pixels, 1r, 1b, 1g, 1k plus a rectangular shape w besides them.
 
because that's all the color information you're saving with 4:2:2 color subsample: one full color signal for every two pixels - and having more white pixels helps DR and high-ISO - but yes, seeing how great the 5D3 is in low light, it could have a few more colored photosites (turn one white to green, an another to black, for example)
 
I was thinking about how would the K (grey ND pixel) show color because the k pixels image would be showed when the rgbw pixels are clipping, so there will be no color information to show a color image and my conclusion is the k pixels could not deliver a color image to show the high lights area. I think to show the high lights area there must be RGB pixels with some ND filter on them or use different shutter speeds in a group of pixels.
 
that's an issue for the debayering algorithm
usually, it will use color from neighboring pixels that are not 100% clipped, and mix that with pure white
yes, it may lead to artifacts in some situations, but if done properly it should be better than a hard white electronic clip
 
Apefos Adapter said:
The pixel count of 5D Mark III is very inteligent and smart. For 16x9 video it will be 5760 x 3240 pixels.
5760 = 1920 x 3 and 3240 = 1080 x 3
So probably the camera will take one pixel from each 9 pixels block. This will probably avoid moire/aliasing and increase resolution.

If Canon could develope a 5D Mark IV with Dual Digic 5+ processors it could do HDR video using the single central pixel from 9 pixels block to do one exposure and use the 8 surrounding pixels from the 9 pixels block to do the second exposure. Using more pixels to get the information of a single pixel can generate more electrons than using just one pixel, so this can be used to do HDR video. This way there is no need to use different shutter speeds to get the two different exposures. Also second exposure using the 8 surrounding pixels from the 9 pixels block could use 2 or 4 or 8 pixels so it could do a 1 or 2 or 3 fstops different exposure when comparing to the exposure from the central single pixel. Both images would be combined in a wider dynamic range image.

HDR video straight from camera is the big step improvement filmmakers are waiting for.
Click to expand...

Well I guess we'll just have to wait till 2015 for Canon to disappoint us again.
 
NormanBates said:
that's an issue for the debayering algorithm
usually, it will use color from neighboring pixels that are not 100% clipped, and mix that with pure white
yes, it may lead to artifacts in some situations, but if done properly it should be better than a hard white electronic clip
Click to expand...

The problem is all the rgb pixels will be clipped in a highlight area and there will be no color information to the debayering algoritm work with. there will be only the k pixel information and it is not possible to extract color just from grey pixels.
 
yes, I know - still, it will be close to pure white, so I think it shouldn't be that hard to make it look better than just clipped 100% white (which is what you get in the situation you're describing) - I have never looked deep enough into a debayering algorithm, though, so I clearly don't know much about this
 
I think the most easy and functional way to do hdr video would be set different iso for the sensor pixels. this way one pixel would be set to 100 iso and the pixel near it would be set to 800 iso. the two different iso could be user defined. two video streams would be captured at the same time, with same shutter speed and same fps and merged in a hdr stream and saved in the memory card. the sensor needs to be a minimal 3840 x 1080 or 1920 x 2160 to allow two 1080p streams at same time, but today most of sensors are more pixels than this, so it would be not so difficult to implement.

There are some similar solutions like this being used today by digital photo cameras, some photo cameras uses different iso to correct dark shadows, lens vignetting and get bettter highlights, so I believe this way to do hdr video will be the way to go.
 
What Canon needs to do is to improve the HIGHLIGHT TONE PRIORITY function. Today this function helps get better highlights but just one fstop difference. Canon needs to enable the option to chose 1, 2, 3, 4 or 5 fstops difference when Highlight Tone Priority is enabled. I think this is not difficult to do.
 
You must log in or register to reply here.
Back
Top