Question to Codec Experts

[jangofettbr]

I was thinking about this and could not have a conclusion so I'm here to ask.

Its possible to elabore a codec that 'encrypt' and 'decrypt' a video so it can bypass copyright infringement?

Like a encrypted video with their image pixels that seems all random and that can be uploaded to youtube for example, without their AI flag it as someones property, and them I use a codec on my site that get this encrypted video and decript in real time as a streaming video.

Using the youtube bandwidth and my site only to parse it as decrypted.

Somebody already thought about this and made it real?

[davexnet]

NO. Youtube re-encodes every thing so it has to be able to properly decode it.

[jangofettbr]

Youtube re-encodes every thing so it has to be able to properly decode it.

I see, its possible using only the image as the parameter?

Get the image and flip it horizontally and vertically, and to parse on my site it make the reverse.

But instead only flipping it, it change the pixels on the image and reverse it to decrypt.

[davexnet]

In theory that may be possible. You could post an unlisted video to your channel to try it.

I suspect google has a way of detecting this kind of manipulation, but I don't know for sure

[jangofettbr]

In theory that may be possible. You could post an unlisted video to your channel to try it.

I suspect google has a way of detecting this kind of manipulation, but I don't know for sure

There is sometimes live pay-per-view content being transmited only changing the image of place or flipping it, now imagine alot of pixels out of order, that even you watching it can't know whats its about.

I think if the order of the encrypted pixels was the same to all videos they could detect it basically with reverse enginering, but with each video having it like a unique hash, that only the site have to decrypted it back, I can't think of a way they can detect it to all of them.

They have to find the video ID specifically being used to delete it manually.

[jagabo]

It's easy to defeat youtube's copyright video detection. The problem is the video becomes too ugly to watch. (I think you also have to distort the audio too now.)

Here's an example from several years ago where someone applied a big watermark:
https://forum.videohelp.com/threads/378783-Removing-white-haze-from-I-would-just-say-s...os#post2447024

[Attachment 84115 - Click to enlarge]

I downloaded the video and was successfully able to remove the watermark with an AviSynth script:

[Attachment 84116 - Click to enlarge]

I believe that video has since been removed. I don't know if they check for that type of "encryption" or if it was reported or manually detected.

[jangofettbr]

I believe that video has since been removed. I don't know if they check for that type of "encryption" or if it was reported or manually detected.

I think the auto detections are flagged immediately and after if someone report it as their property.

The problem is the video becomes too ugly to watch.

When I mean its encrypted to change the image pixels, its something like this:

[Attachment 84117 - Click to enlarge]

If you watch it on youtube its nothing but random pixels, but then its "rearranged" or "decrypted" to a normal image again on my site.

[jagabo]

After you upload to youtube the video will be re-encoded. The video you download later will be different than the one you uploaded. It will not decrypt properly. You'll need a much more sophisticated algorithm that is insensitive to losses.

[Cornucopia]

One thing you may not be considering is that codecs get some of their efficiency by analyzing consistency, similarly, contiguous areas, etc.
What you intend (rearranging the pixels) is equivalent to adding a bunch of random noise to the image, and we all know that random noise is the WORST thing for a codec to work with. The resulting bitrate would be ridiculously high compared to what you would normally expect from that codec with that material.


Scott

[jangofettbr]

After you upload to youtube the video will be re-encoded. The video you download later will be different than the one you uploaded. It will not decrypt properly. You'll need a much more sophisticated algorithm that is insensitive to losses.

I have tested it with little pixel noises and a minimum loss compression get the image back but with a lower quality.

I think the only way its possible, its through scrambling by blocks, with larger pixels:

[Attachment 84122 - Click to enlarge]

Gonna try it later.

One thing you may not be considering is that codecs get some of their efficiency by analyzing consistency, similarly, contiguous areas, etc.
What you intend (rearranging the pixels) is equivalent to adding a bunch of random noise to the image, and we all know that random noise is the WORST thing for a codec to work with. The resulting bitrate would be ridiculously high compared to what you would normally expect from that codec with that material.

The consistency could be applied in the images after its decryption.
It decrypt back to the normal image and then the encrypted ones are removed from the cache.

Its all hypothesis from my head, only conjuring if its possible or not.

[jangofettbr]

CONCLUSION:
Its possible but still get little artifacts from the compression and changed the brightness a bit.

The image was scrambled by blocks and uploaded on youtube, then downloaded and rearranged.


[Attachment 84145 - Click to enlarge]

[Attachment 84144 - Click to enlarge]

[_Al_]

Its cool. What are you using for slicing, opencv, vapoursynth or something else?
Also are you uploading higher resolution to youtube, where some blocks are just fillers or maybe those posted images in posts #10 and #11 are not related ?

[jangofettbr]

Its cool. What are you using for slicing, opencv, vapoursynth or something else?
Also are you uploading higher resolution to youtube, where some blocks are just fillers or maybe those posted images in posts #10 and #11 are not related ?

Not related.

I found this python script that "encrypt" the image by block-scramblying and it uses Pillow:
https://github.com/snekbeater/scrambpy

It have a option to double the size in the output of the encrypted image, then decrypt to its normal size, thats what I did with a image of 1920x1080.

Tried with a noise encryption before, and it works but the images are more affected by the compression.

Only testing out of curiosity

[_Al_]

Oh python, for whatever thought/idea, there is already a module for it.