I am at my witts end trying to reverse MPEG-4 compression. Help!!!

[forensic]

Does anyone know of any software that can reverse the blockiness caused by MPEG-4 compression used with security DVRs?

The DVR compresses the video on the fly. At some later time, a segment of that video is exported as an uncompressed AVI file. However, the substantial loss of quality from the original compression is quite evident.

I need software, or a plug-in, that can mathematically reconstruct some of that lost detail by reversing the formulas originally used in the compression, likely by using adjacent frames. I don't care if the software is slow or pricey, but it does need to work.

I have tried various deblocking filters with VitualDubMod and AviSynth scripts, but the end result simply softens where the blockiness meets. Becuase of the artifacts, attempts to Super Resolution or frame average from there produce no increase in resolution.

It is not a matter of codecs as I can open the uncompressed AVI file and save it as whatever I want. What I need is software or a filter to mathematically reverse the original DVR MPEG-4 compression process as much as possible.

The attached test AVI is compressed with LAGARITH lossless codec to so it could be posted. I truly appreciate any guidance or solutions you may have.

P.S. What if the DVR used Wavelet compression instead?

test.avi

[lordsmurf]

Contrary to what television shows like CSI might show, such a task is impossible. Sorry. You can't make mousse from a dog turd. Garbage in yields garbage out.

[jagabo]

Here's a list of 8 numbers:

100 0 80 20 60 40 50 50

Lets compress it down to 4 numbers by averaging pairs together:

50 50 50 50

Now, given only the list of 4 numbers, reconstruct the original 8. Having trouble?

[guns1inger]

Contrary to what television shows like CSI might show, such a task is impossible. Sorry.

Are you trying to tell me that television lies ?

Say it ain't so

[redwudz]

I'm a bit confused with this:

At some later time, a segment of that video is exported as an uncompressed AVI file.

The DVRs that I have used export the video in the same format as it's stored on the unit, most times to a USB thumb drive.

No help with the present problem, but for future use, a higher resolution camera or DVR setting might be more appropriate if you want to recognize individuals or license plates. I assume that video was cropped down from a larger frame or it's a long telephoto. Depending on the cost value of the incident, an upgrade of the equipment might be cost effective. I either use more cameras or a telephoto camera aimed at a key location. It doesn't do much good if your video can't identify much more than the make of a vehicle and the time of the incident.

For an enclosed facility, I use a secondary camera at the entrance gate that is close enough to recognize individuals and license plates. Checking the time codes against the second camera is enough to show they are one in the same.

I use this company a fair amount: http://www.supercircuits.com/index.asp

[forensic]

While I agree with, and appreciate, the challenges of undoing the compression. Is there anything worth trying that may be better than deblocking? Especially since we have adjacent frames.

[guns1inger]

I have not seen anything that does what you ask. It is quite a challenge you have set for software to achieve

1. Identify artifacts
2. Scan backwards and forwards looking for matching sections of adjacent frames that do not have artifacts, or have reduced artifacts
3. Rebuild frames from collected data

There is software that does rebuild frames from adjacent frames. Mokey is one. It is used for wire rig and object removal from frames, and can be very effective. But it isn't for restoration of over compressed data, and I can't see any way to use it for that task.

The final thing to note : if this is coming from security cameras and you are looking to use to identify an offender, for instance, and you tamper with the footage, the defense will have a field day in court. How can you prove that any reconstructed footage relates in any way to reality ?

[AlanHK]

Does anyone know of any software that can reverse the blockiness caused by MPEG-4 compression used with security DVRs?

The DVR compresses the video on the fly. At some later time, a segment of that video is exported as an uncompressed AVI file. However, the substantial loss of quality from the original compression is quite evident.

I don't understand. How do you know that there is a "substantial loss of quality from the original compression"? If you can view it, why can't you use (capture, if no other way) the "original" version?

Otherwise, Super-resolution may be able to combine information from several frames to deduce details. But I wouldn't try to introduce such "enhanced" video as legal evidence.

[olyteddy]

... Sorry. You can't make mousse from a dog turd. Garbage in yields garbage out...

Yeah, as http://www.utahphillips.com/utah.html teaches us you need a Moose Turd to make Moose Turd Pie... Seriously though, I wish it were possible.

[koopak]

Hmm, super-resolution techniques might well still apply.

As for the 50 50 50 50 example, although I appreciate the sentiment, we do not live in a world of white noise, but where there are discernible and predictable patterns, and so given a mean value and an estimated curve, you can reconstruct information.

For instance:

10 - 20 - 30 - 40 - 50 - 60 - 70 - 80 - 90

10 - xx - 30 - xx - 50 - xx - 70 - xx - 90

Now if you had to "guess" only from the second line what was in the first, could you? As that part of the image might represent a smooth gradient, you would probably guess right!

And to take the example one step further, if the next frame has:

xx - 20 - xx - 40 - xx - 60 - xx - 80 - xx

Then you're laughing, because provided there is no motion in that area, you will be able to perfectly recover your information loss.

Anyway, as far as software that can do this for you, I strongly recommend you check out a company called "Salient Stills" and ask them for a demo of their program. They sell some of the tools the FBI use to do CSI-like things. Their 'VideoFocus Pro' suite is the reason the mugshots they publish from 320x240 Quik-E-Mart $2 CMOS 'security cams' (*cough* webcams *cough*) actually end up with convictions!

Good luck!