Problem with Canon 5D Mk. II video?
 

It seems there is a serious problem with the video 5D mark II.

On the Russian site, we found abridged level's range of video, compared with the same photo.
Analysis was made on Vincent Laforet' material:
- Video: Canon Digital Learning Center - EOS 5D Mark II: Full-Resolution Video Clips
- Photo: Canon Digital Learning Center - EOS 5D Mark II: Sample Photographs

Let us replace the 100% black to red, 100% white to blue:
- Original photo (excellent result): http://s52.radikal.ru/i135/0809/11/c8301fef0328.jpg
- Original photo with the levels cuttings down to 16-235: http://s60.radikal.ru/i170/0809/1a/9ea602ca9bef.jpg
- Screenshot of the video (very similar to a photo with the levels 16-235): http://s39.radikal.ru/i085/0809/98/cbf133ce4767.jpg

The material is taken from Russian Forum: Ôîðóì VIDEOMAX: Canon EOS 5D Mark II - ïîääåðæêà 1080p30,38.6 mbps

The problem even more serious.
Made an assumption that the processing of video before recording to the media is as follows:
- Takes the original picture (histogram): http://s46.radikal.ru/i112/0809/96/35469b3d67c2.jpg
- Cut levels down 16-235: http://s61.radikal.ru/i172/0809/8e/aee1573f4285.jpg
- Stretched levels to 0-255 breaks the histogram: http://s59.radikal.ru/i163/0809/64/dbf3c1bd6507.jpg
So this algorithm appears in After Affects.
In fact, the processor 5D mark II breaks the histogram by a nonlinear function: http://i060.radikal.ru/0809/25/7c31826d55f6.jpg

Source: Ôîðóì VIDEOMAX: Canon EOS 5D Mark II - ïîääåðæêà 1080p30,38.6 mbps

Anybody can repeat the analysis and confirm these results?

I think I may be lost Ivan... are you trying to say the video has been processed in some way?

No. It was not processed by Vincent Laforet (or anybody else).

Yes. It was processed by 5D mаrk II, during writing video on media.

Ivan,

I can't read Russian, so I'm wondering if they explained how they obtained the video "screenshots". Screen capture or frame export in Quicktime Pro? Not sure we can't rule out another factor here.

It was used After Affects.

The result was repeated in:
- MPEG Streamclip (grab saved as tga) and Adobe Photoshop CS3 (histogram): Ôîðóì VIDEOMAX: Canon EOS 5D Mark II - ïîääåðæêà 1080p30,38.6 mbps

- RGB-analysis in FastStone: Ôîðóì VIDEOMAX: Canon EOS 5D Mark II - ïîääåðæêà 1080p30,38.6 mbps

- analysis of .mov-file in Adobe Photoshop CS3: Ôîðóì VIDEOMAX: Canon EOS 5D Mark II - ïîääåðæêà 1080p30,38.6 mbps

Gamma Curve
 

Of course its non-linear. The RAW data is 12-bit or perhaps 14-bit in terms of the Mark II. The codec saves 8-bit. The mapping between the 12-bit / 14-bit to 8-bit is done via gamma curve.
For film or the cine-modes in most prosumer cameras, there is a non-linear curve in order to extract highlight information.
Gamma Curve profiles can be loaded onto the 5D if I'm not mistaken. You dont want a linear curve, you want an S-shaped curve for compressing highlights and extract more highlight information...

what Ivan wants to say :
Comparing a picture taken from still and video, it seems the processor is goes pretty easy on compression. by just cutting values where it should rescale them.
and the calculations are pretty simplistic too.
I do not think it is a problem if you do not compare stills to video.
If use this device like a camera, consider only the video output.
if you use it as a still camera, for sure you will set up differently.

Yes.
Thanks, Giroud. You help me very much.

<<< the processor is goes pretty easy on compression >>>
E.g. camcorder Sony Z1 uses 14-Bit HD DXP (Digital eXtrended Processor) and writes 8-Bit video on media. Histogram is smooth without breaks.

5D mark II converts 14-bit signal into 8-bit video with breaks in histogram. We are losing in gradations.

Tyler
copy the link from the web page, then open a new window to google.com. paste the link in the SEARCH box, then press ok. google will give you an option to translate the page for you.

ken

This is not a problem. Most all video cameras do this as recording more than 8-bits of data is very hard to do. There are very few video cameras that record more than 8-bits of data. That's the reason DSLR can shoot faster and longer with 8-bit jpegs vs. 10 or 12 bit RAW files.

Yes, most of all video cameras recording 8-bits of data.
The problem with D5mkII - it recording 8-bits with breaks in histogram. Others - without such the breaks.
Look at Adobe Photoshop CS3 histograms:
- 5D mark II: http://www.videomax.ru/forum/uploads/16534/61.jpg
- Nikon D90: http://www.videomax.ru/forum/uploads/16534/68.jpg
D90 histogram is smooth.
5D histogram with breaks - thin gradations is lost.

It's seems as 5D processor's bug.

Actually, it's not the camera that's doing this - it's AE. The camera clearly maps everything to 16-235, which puts it into the NTSC legal levels. AE looks like it's stretching it back to the full 0-255. You can get dramatically different results by switching your working color space - ProPhoto RGB looks to be the correct working color space:

Yep!
It was checked in various ways. Exported from Quicktime Pro, and directly imported in Photoshop CS3.

I wasn't familiar with ProPhoto RGB but found a good article on it here:

Understanding ProPhoto RGB

It's basically got a much wider color gamut than the Adobe or sRGB color spaces, which makes it a good choice to preserve as much color information as possible when converting the RAW data from the sensor into 8-bit RGB for video.

BTW: ProfotoRGB works well only 16bit mode. In 8bit mode may cause problems like solarization etc.