|
|
| DougC |
| Posted: Feb 21 2008, 04:10 PM |
 |
|
Newbie
Group: Members
Posts: 1
Member No.: 23100
Joined: 21-February 08
|
Low frame rate and high compression are a bad combination for video enhancement. Yet that is what our police department's crime lab is working with. We have tried some deblocking and noise reduction filters with no noticeable improvement in either the suspect's face or the static background. Albeit, we are not very skilled at this software or filter settings.
A MPG4 test video is located at http://www.protectyourplace.com/test.avi
A alternate test video is located at http://www.protectyourplace.com/test2.avi
Does anyone have the skills to know what filters (with what settings) and in what order will provide useable clarity? If we can apply your techniques to our unsolved cases, you will earn the bragging rights to helping us solve some high-profile crimes. I will check this forum regularly and can be emailed directly at "forum" at the domain "ProtectYourPlace.com". Thank you, DougC
EDITED ADDITION: I do not have the original. Besides, the goal is to see if it is possible to enhance the test video as it exemplifies the typical security camera video. If it can be done, then the same technique can be used with other unsolved crimes. The best filter/process sequence/solution will be presented to the forensic departments and the forum contributor will be acknowledged in writing (a great addition to anyone's resume) if you want the credit.
I have tried the suggested MSU filters, as well as the "neuron" and "bob" filters, and found no significant benefit to my test video. I don't care if the process is ugly...just that it gets to the finish line. We have a lot of unsolved murders and child abductions because of poor video quality. Any help would be greatly appreciated.
FURTHER ADDITION: SPM-MAD & ARKLIGHT, I really appreciate your help.
Color correction after, spliting channels did not help with this black/white video, nor the test2.avi color video. Likewise, almost every combination of filters is getting me no where special. Would I see dramatically better results with the high priced commercial software? Why can't I get any improvements with the static background? I tried deblocking, denoise and smoothing. |
|
| Loadus |
| Posted: Feb 21 2008, 07:24 PM |
|
|
Advanced Member
Group: Members
Posts: 352
Member No.: 10881
Joined: 1-July 04
|
Can you provide the original size interlaced video - that would have more visual information on it,
although that is pretty difficult to enhance (not impossible, just difficult). 
--------------------
deviantART behance |
|
| -SPM-Mad |
| Posted: Feb 21 2008, 11:15 PM |
|
|
Advanced Member
Group: Members
Posts: 197
Member No.: 1332
Joined: 13-December 02
|
Dear DougC,
I read your additional information that you do not have the original video right now. But as Loadus said, interlaced video contains twice the temporal information (and half the resolution) of progressiv video. Simple deinterlacer interpolate the resolution and throw away half of the temporal information which is not the best way.
A filter that preserves both information is for example: http://neuron2.net/bob.html
There are several other filters that could enhance the movement but usually just by means of interpolation which does not add any real picture information. So any other adjustment I suggest are only for color/brightness.
A set of useful filters for some quality enhancement are the MSU filters:
http://compression.ru/video/denoising/index_en.html
http://compression.ru/video/deblocking/index_en.html
http://compression.ru/video/smart_contrast...t/index_en.html
http://compression.ru/video/deblocking/sma...locking_en.html
But the results promised on the website I usually only get in very rare cases. Often the improvement will be only subtle.
In the example video you provided I used the MSU Deblocking filters and added a 'levels' filter with Black: 41, White: 198, Middlepoint: 0.785
I also tried playing around with smart Smoother (http://neuron2.net/hiq/smoothhiq.html) and 'temporal smoother', both which help to reduce noise. But that does not help with the faces of people moving.
If there is non-static noise over someones face who is NOT MOVING, then the 'temporal smoother' with a high value helps alot. When the people move through the picture it does not help.
Hopefully I managed to give you one or two ideas,
Greetings
-SPM-Mad |
|
| arklight |
| Posted: Feb 22 2008, 06:51 PM |
|
|
Advanced Member
Group: Members
Posts: 135
Member No.: 8316
Joined: 16-January 04
|
Im a bit ill, excuse me if I don't go in too deep
but let's have a little try...
here's an original frame from the footage....
Here's the "virtualdubbed" enhanced frame of
the man.....
Not much done too it apart from good old fashion resizing,
then a framegrab, then used curves to seperate dark, lights
and mids so features stood out...
you could REALLY go to town if you have a node based
compositor and split the picture into seperate
grayscale colour channels (usually alot of info in the green channel)
then recombine (or leave it at one
channel since it's nearly grayscale anyway)
since you want to see features and not lovely colours.
Oh yeah, it's blurry a bit. Could sort that out,
resize in steps etc, extreme sharpening seemed to
create lots of artifacts.
But if you stand back from screen,
you can see his features.
Heri |
|
| -SPM-Mad |
| Posted: Feb 22 2008, 06:59 PM |
|
|
Advanced Member
Group: Members
Posts: 197
Member No.: 1332
Joined: 13-December 02
|
The picture from the camera is grayscale anways. Any color-information we see in the video is from analog signal-noise.
So by isolating the differences in the colorchannels from grey and substracing this from the final video we might improve it. I have not tried fiddling around with that, yet.
-SPM-Mad |
|
| arklight |
| Posted: Feb 22 2008, 08:35 PM |
|
|
Advanced Member
Group: Members
Posts: 135
Member No.: 8316
Joined: 16-January 04
|
yeah you r right SPM!
tis grayscale,i'm ill, saw the color fringing/strobing
thought it was just poor quality cctv!
As for the finding/enhancing information
in an image look at this video tutorial....
http://library.creativecow.net/articles/sh...eo-tutorial.php
Yes, it's about keying DV, But it's how he
get's there that was intriguing for me
and i've applied similar solutions to other image
problems...
I don't think we have any kind of hard core color
seperation and splitting/combining.
split/unsplit channels...?
all i could find was "unsplit/split channels"
where you split channels in the filter stack, then
use levels, whatever, to alter them, then add
unsplit channels to combine the result.
couldn't find a direct link for original filter,
just this link as part of a virtualdub filter pack...
http://altukhov.clan.su/forum/20-245-1
i recommend you scan that pack first coz,
it's not a familiar place to me.
for those who don't know how to use filters,
virtualdub newbies...read this...
http://forums.virtualdub.org/index.php?act...ST&f=4&t=14396&
heri |
|
| arklight |
| Posted: Feb 28 2008, 07:55 PM |
|
|
Advanced Member
Group: Members
Posts: 135
Member No.: 8316
Joined: 16-January 04
|
There's the other test cctv sample to try out.
Here's the original,framegrab at original frame size...
Here's the original,framegrab at original frame size...
virtualdubbed version....
virtualdubbed,alt bigger version....
used mainly a "grain technique" discussed at another forum...
http://neuron2.net/board/viewtopic.php?p=7...7afe6efdb468e69
But i'm sure there are a thousand ways to do it,
probably much better I just had a quick hack at it.
Grain technique....
grain tchnique is basic, get emilano's add grain filter,
it's in the pack here....unzip pack into plugins folder..
http://emiliano.deepabyss.org/
Rez up as high as your computer can handle,
I did 2000 percent resize( nearest neighbour),
then added grain, then after added smart resize filter
(resize down to 720x576 precise bilenear comes wirth blur)
so in order filter stack should be in this order
.1 smart resize filter/any resize filter(nearest neighbour high as you can rez up)
.2 add grain
.3 smart resize filter( to 720x576-PAL size)
Go back into add grain filter, check rgb, and gaussian,
then mess about with the intensity and
distribution of the grain till
features become apparent.
I then did a frame grab.
got it into an image software...opened up smartcurve,
used curves to "ummuddy" the colours and lighting
so features stood out.
of course better source! Better results.
I'm sure there are other ways of doing this
problem.
Heri |
|
| arklight |
| Posted: Jun 1 2010, 07:08 PM |
|
|
Advanced Member
Group: Members
Posts: 135
Member No.: 8316
Joined: 16-January 04
|
I was flicking channels when I saw it...,,exactly what we've
been talking about in this thread being applied.
It was one of those Police pron shows with clips of
car chases and crims on the UK's Channel FIVE i think.
And in half of the cctv being shown, especially when
they cut to a close up of a pereps face within the
cctv footage. There was the tell tell grain , from the
future is grain uprezzing techniques developed years ago
Yes, that's what bought me back to the boards! Nearly fell off my chair.
nicely done though.
Heri mkocha |
|
