How to say for certain which BT coefficients a video uses?

[fvisagie]

Hi All,

I'm working on a project with multiple sources and need to get the colour handling right. From some devices I have clips in multiple formats, e.g. some Nokia N8 AVC clips are 640x480 and others are 1280x720. Not having found any specifications on the matter, the question is whether the manufacturer decided to always use the same coefficients, or to use coefficients according to resolution.

MediaInfo only goes so far as saying YUV 4:2:0. In AvsPmod the SD ones have a tiny bit more contrast and saturation with the preview display set to Rec.601, and likewise for the 1280x720 clips at Rec.709. So it seems the manufacturer uses coefficients according to resolution, but, is there any more scientific or exact way of determining which coefficients a particular clip uses?

Thanks a lot,
Francois

[poisondeathray]

It's impossible to say , unless you shot a test pattern (e.g. color bars) at the beginning of the clip

Some cameras record metadata that say which matrix was used

Otherwise it's safe to assume Rec601 for SD, 709 for HD

[fvisagie]

It's impossible to say , unless you shot a test pattern (e.g. color bars) at the beginning of the clip

Thanks for the suggestion, poisondeathray. I guess I could always make such shots after the fact, ensuring they're done at the corresponding camera settings. Where can I find a suitable test pattern for this? And is there anything to doing this other than displaying the results in both Rec.601 and Rec.709 and see which more closely matches the input?

Edit: Or, by mentioning color bars are you implying that something as simple as ColorBars() would suffice?

[poisondeathray]

But how you you get your camera to see colorbars() ?

But you could do it that way , point the camera at something and just see which more closely matches with different matrix when converting the footage to RGB for display.

There are professional test charts and cards availble (expensive)

There are 709/601 test patches on the free belle nuit chart, but not sure how you would get your camera to see it (you can use it for software testing 709/601 conversion testing) . Maybe put it on the HDTV and point the camera at it
http://www.belle-nuit.com/test-chart

You can convert to YUY2 and use videoscope() to see if the luma values are the same in the 709/601 grouping

[poisondeathray]

If you're going to use that test chart, make sure you convert it correctly or keep track of the various YUV<=>RGB conversions

eg. if your TV has a bmp reader (e.g. off usb), convert the tiff to bmp and it should be RGB . But if you convert it to YUV, then you have to be sure how to convert it to YUV and what matrix your display equipment using to convert YUV to RGB.

If you've calibrated all your equipment, and are certain that 709 is used in a certain pathway e.g. blu-ray , then can make a blu-ray test image disc

[fvisagie]

I've managed to find what seem to be the official Nokia N8 specs, but unfortunately they don't specify ITU-R output formats.

In the meantime, I'm doggedly working through your suggestions, thanks . I'd also imagined pointing the camera at a test display as you mention; which display could also influence results of course.

[fvisagie]

That's a really useful test chart, thanks. Unfortunately my display's limitations meant that I couldn't produce reliable/meaningful test input for my camera with it.

Fortunately however, the weight of evidence does point to BT.601 being associated with SD and BT.709 with HD, specifically also with this camera:

• Its 'video/h264' encoder output formats seem to be associated with AVC recording; the 'video/mp4v-es' formats not
• Comparing test encodes shows that the camera does produce AVC encodes at HD settings and ASP at SD
• Sources like this one have it that AVC is associated with BT.709 and ASP with BT.601
• Plus, the HD encodes do look better with AvsPmod set at BT.709 and the SD ones at BT.601

So I'll do my best to convince myself that this issue is now put to bed .

VideoScope() is a very useful tool, by the way. I was able to confirm its results with the Belle Nuit chart by reading off the RGB values shown in AvsPmod's status bar and converting those to luma with the formulas given on the chart's web page.

Thanks for your help!

Francois