I've got half a dozen security cams. Some are
straight IP, others - PI and others - use 'motion'
to create an MPEG stream.

Wrote a display app (python) that rolls through
the cams so I can see what's going on. Also a
sep app that RECORDS selected cams as MP4s
mostly using OpenCV stuff.

HOWEVER ... while I do have a little daemon that
sets day/night values for v4l2 that's not always
good enough. Super-bright, or cloudy, days can
result in kinda crappy images.

Tried simple histogram expansion - but that kinda
looked WORSE in most cases - TOO much contrast.

SO ... wrote a little app that breaks the images
into a 7x5 matrix and yields 'average brightness'
figures for each segment ... the basic of 'matrix
metering'.

Alas even after extensive searching I can't find
any advice on HOW to make use of those figures,
HOW to adjust brightness/contrast based on that
matrix analysis.

Camera makers HAVE figured this out, and 'matrix'
almost always yields great images. Alas the HOW
is not published.

OpenCV2 can adjust both brightness and contrast
quite easily - but what VALUES to feed it .....

Have you seen anybody addressing this problem,
this issue, on the net where I might get advice
on how best to analyze those image segments and
derive the needed adjustment figures ??? This
drifts into "statistics" - not my best issue.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On Sun, 4 May 2025 01:24:21 -0400, c186282 wrote:

SO ... wrote a little app that breaks the images into a 7x5 matrix and
yields 'average brightness'
figures for each segment ... the basic of 'matrix metering'.

That's sort of like pooling in a convolutional neural network. The image
is broken into 2x2 or similar and either the maximum or the average for
the square is used for the next layer. It's interesting to look at the
images after each iteration. It really sharpens lines and shapes that
would be a smoother gradient in the original image and reduces the
dimensions.

I don't know if successive iterations would yield something you could use
to set contrast.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On 5/4/25 4:16 AM, rbowman wrote:

On Sun, 4 May 2025 01:24:21 -0400, c186282 wrote:

SO ... wrote a little app that breaks the images into a 7x5 matrix and
yields 'average brightness'
figures for each segment ... the basic of 'matrix metering'.

That's sort of like pooling in a convolutional neural network. The image
is broken into 2x2 or similar and either the maximum or the average for
the square is used for the next layer. It's interesting to look at the
images after each iteration. It really sharpens lines and shapes that
would be a smoother gradient in the original image and reduces the dimensions.

I don't know if successive iterations would yield something you could use
to set contrast.

What I need is advice on how to best CALCULATE from
each segment what the contrast and brightness may
be and then figure out what to DO about it.

A real-world image ... sun is bright or dim, there
are a few really bright PARTS in it and some really
DARK parts. The "best picture takes such stuff
into account without getting TOO extreme in any
direction. Camera-makers seem to have this down
quite well, but their algos are 'proprietary' and
even extensive searching has not yielded much
detail into how they do it (with little micro-
controllers).

Gonna have to work on this for awhile ... do lots
of experiments.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)