Re: [CinePaint] Looking for a feature useful to smooth timelapse sequences - grading

On Wed, 22 Nov 2006 12:53:52 -0800
"John Arrowwood" <[email protected]> wrote:

> 
> I don't want a brute-force approach.  I don't want to experiment on
> every frame to try and figure out what settings will work.  I want to
> mathematically analyze the frame image and the average image, and
> calculate the adjustments that will be necessary in order to bring the
> histogram of the frame into alignment with the average.  Now,
> mathematically, it's not trivial.  The histograms may have noticably
> different shapes.  But the end result should be that if you have a
> field of blue, it should, on average, be around the same shade of
> blue.  If you have a field of white, it should be around the same
> shade of white.  The frame image WILL have more details than the
> average, and that's why the histograms won't match exactly.
> 

The only kind of automated analysis would be to determine
the black point and white point of every image and adjust
the histogram to fit between these two points.  Of course,
an underexposed or overexposed image, with a high percentage
of the detail being either lost in shadow or blown out in
the highlights, will essentially be beyond redemption.
There will be no way to ever bring these back into conformity
with the "average."

> Picture this:  You are given two images.  Your task is to
> AUTOMATICALLY adjust the histogram of the one so that it
> matches as closely as possible the second.

With a time lapse sequence I doubt that this could ever be
possible.  Each subsequent image would introduce new detail
and each successive image would necessarily present a unique
histogram.  (This is why I have questioned the idea of an
"average" of images that are bound to be significantly
different.)

But the goal is not to match histograms, but only to insure
that the intensity distribution encompasses as much as possible
the available space.

There is a technique known as "histogram equalization" but
this effect often leads to unnatural contrast effects and is
usually employed only for scientific purposes.

Frank Peters




-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV