On Sat, 21 Oct 2023 16:51:49 +0100, Vir Campestris
Post by Vir CampestrisIt doesn't matter what you do with the signal if the fundamental problem
is that it contains less information. Making it negative is just turning
it upside down, but it's still based on the amount of light that entered
the lens from the subject, and you won't get any more of it by messing
about with the signal after the event. You can improve matters by
shining more light on the subject, or using a bigger lens (lower f stop)
or a more sensitive pickup device, or designing less noisy video
amplifiers, but there will always be less original information available
from darker surfaces.
That's not the problem.
When he holds still I can see him perfectly well. But when he nods his
head the CODEC only updates the edges of his forehead, while moving
everything from the eyes down and the outline of his head correctly.
After a while (next I frame?) his forehead catches up.
My guess is is that the authors have chosen to reduce the update
frequency on dark surfaces - as you say they aren't as easily seen.
In order for the system to update something that has changed between
frames, it has to detect that something has in fact changed. It can
only do that by comparing information between frames. This will
inevitably be more difficult to do in parts of the frame that contain
less information, so that changes are less apparent.
For me, the type of shot that shows up low bit rates most readily is
in a low key scene where an actor is lit from the side and they nod
their head, whereupon their face almost seems to split into two parts
that move independently. The two sides of the face are at different
brightness levels and so the system treats them differently, the
brighter side presumably receiving more frequent and/or more accurate
updates because there is more information to work on.
Rod.