[RFC] A safe area in steganography to provide high anti-compressibility

[SunriseFox]

Metadata

Field	Value
Manifest	Meta/Article-3
Purpose	provide high anti-compressibility
Feedback deadline	2019-11-23 12:00:00 UTC
Explicitly designated reviewers	@neruthes @guanbinrui

Abstract

We cannot prevent facebook/twitter from compressing our images and causing data loss. As the current implementation of steganography relies heavily on the raw original data of images, compressed images may not be acceptable for reading data.

Attempt to download all original images and decrypt is neither a good idea. It consumes too many bandwidths of users, as well as slows down the decryption progress.

I think we could provide a safe area on the image, e.g, a corner. In this area, we encrypt some metadata with high anti-compressibility (it is possible, such as barcode). Only after successfully decrypted the metadata and verified it shall we try to decrypt the main part of the image. If failed, try fetching the original image and try again.

Dependencies

#354

Glossary

Phrase	Definition
...	...

Notes

• I can hardly decrypt a message with an encrypted image.

[neruthes]

We can use the top-left 25vw*25vw corner to do this. In this area, the pixel blocks should be larger. So that only if we can detect a pattern in this corner will we try downloading the whole image. Nice idea! 👍

[neruthes]

The easiest pattern may be a 4*4 black-and-white chess pattern, with top-left being black.

[guanbinrui]

I think it works when vendor gives user highly compressed image in order to reduce using of bandwidths. We can add the safe area to the base image and create a coressponing mask image which hide the area from steganography.

[neruthes]

[Jack-Works]

Objection, this make steganography no longer "steganography"

[neruthes]

@Jack-Works I suggest that hiding the fact that an image contains a Maskbook payload is not necessary at this stage. There are many months ahead before we need to introduce countermeasures against anti-Maskbook censorship.

[Jack-Works]

@Jack-Works I suggest that hiding the fact that an image contains a Maskbook payload is not necessary at this stage. There are many months ahead before we need to introduce countermeasures against anti-Maskbook censorship.

If the design target is not tending to prevent anti-Maskbook, please change a term, use "steganography" may mislead users and they will think this "steganography" feature can keep them safe.

[SunriseFox]

It stays safe. We just increase the anti-compressibility. We do not expose anything new to anyone.

[neruthes]

@Jack-Works I suggest that hiding the fact that an image contains a Maskbook payload is not necessary at this stage. There are many months ahead before we need to introduce countermeasures against anti-Maskbook censorship.

If the design target is not tending to prevent anti-Maskbook, please change a term, use "steganography" may mislead users and they will think this "steganography" feature can keep them safe.
— @Jack-Works

Image-Based Payload (IBP) might be a good option.

It stays safe. We just increase the anti-compressibility. We do not expose anything new to anyone.
— @SunriseFox

Yes. Allowing only one base image is already non-confidential enough. This pattern does not introduce new risks when it enhances recognizability for Maskwork-compatible softwares.

[neruthes]

I think we can ratify this now.

[neruthes]

I will mark this as Ratified at today 05:13 UTC if no further discussion.

cc @yisiliu @Jack-Works @SunriseFox @guanbinrui

[guanbinrui]

For now we will judge image payload dimension before decoding it and there is no ORIGINAL IMAGE we can found. So it may not urgent for this?
Futher more the mosaic style metadata really vision disturbing. What about use the MASK FACE for this purpose?

[yisiliu]

Objection, this make steganography no longer "steganography"

In our current version, we are posting a Maskbook poster, which is already too open to hide. When we decide to go full steal mode, then it's the time to consider removing all "public" features from the image.

[Tedko]

Mask face is a good idea

…

On Tue, Dec 10, 2019 at 12:25 guanbinrui ***@***.***> wrote: For now we will judge image payload dimension before decoding it and there is no *ORIGINAL IMAGE* we can found. So it may not urgent for this? Futher more the mosaic style metadata really vision disturbing. What about use the *MASK FACE* for this purpose? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#410?email_source=notifications&email_token=ABTAVTJWXK2MBXMO2YCH2HDQX4K3BA5CNFSM4JP5H3J2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEGNFGRY#issuecomment-563761991>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABTAVTOJHSC4PBBRMGZSLODQX4K3BANCNFSM4JP5H3JQ> .

[yisiliu]

Mask face is fixed in the middle, being a good option for our validation purpose.

[neruthes]

The chess board should be avoided if possible.

[guanbinrui]

As the disscusion above the chess board opinion will be considered only if we can't get stable validation result from MASK FACE and I think we can mark this issue as Ratified if no further discussion.

@SunriseFox @Jack-Works @yisiliu @Tedko @neruthes

[yisiliu]

Let add this to our steganography process:

• Use the mask face as the validation area
• If it doesn't work, try at most 3 times
  • If it works, go ahead
  • if it doesn't, fall back to url based payload

[Jack-Works]

It stays safe. We just increase the anti-compressibility. We do not expose anything new to anyone. -- @SunriseFox

The ability to distinguish if someone is using steganography is a serious information leak. (though preventing this type of leaking is not our goal at the current process accroding to @neruthes )

We should really change the term steganography to something else to avoid the misleading to user since we're not real secure at the steganography level. (Like Image-Based Payload by @neruthes )

[guanbinrui]

The ability to distinguish if someone is using steganography is a serious information leak.

Steganography do not mean 100% secret. Beside steganography exists steganalysis detecting messages hidden in image using steganography.

But I think Image-Based Payload is a more user friendly name.

[Tedko]

Any updates on this? Let’s use the mask as an indicator etc.

[neruthes]

I think that the term Image-Based Payload may be used in all contexts to address this feature.

[neruthes]

But wording is beyond the scope of this issue. Please talk in alternative places.

[Tedko]

@guanbinrui @SunriseFox I think we can close this issue? Seems like the Image-Based Payload decryption is working well