Originally Posted by JohnAd
So around 10% of frames have peak white over 235 with the highest at 252 at frame 3005
Thanks for the numbers. I will follow with some images...
Originally Posted by Mr.D
Not sure I follow you. There should be zero clipping going on in the chain up till the display. At this point we are really only concerned with where to clip white in RGB referenced to the end display itself.
Yes, but if we find that our sources exceed 16-235 only in RGB but not in YCbCr, we can improve the constrast by changing the YCbCr->RGB conversion.
Using mpc-hc, avisynth, and 3 custom 3DLUT files I've created, here are some images to show how much <16 and >235 data exists in a movie.
I've used it with the first 1m20s of the Blu-ray Babylon.
The zip file I linked contain 2 frames of the movie. One with a near reference black image (Babylon007), and another with a near reference white image (Babylon011).
The images without any extension are without any masking.
The images with "RGB_16-235_GR" in the name show pixels with RGB < [16,16,16] at green, and pixels with R>235 or G>235 or B>235 at red.
The images with "Y_16-235_GR" in the name show pixels with Y < 16 at green, pixels with Y>235 at red, and pixels with CbCr outside the [16-240] range at yellow.
As we can see in image 7, there are a lot of pixels < 16 in both RGB and YCbCr, but when we look at the higher level side (using image 11), > 235, eventhough there are a lot of pixels with R,G or B > 235, there aren't any with Y >235...
The images are PNGs at native resolution, so I zipped them and uploaded the pack here