Too many wet positions in Wet-Paper Codes steganography with random H

Martin Benes

10/14/23, 12:08 PM

I am implementing wet-paper codes (WPC) with randomly generated parity-check matrix $H$, based on this paper. As the wet DCT coefficients, I set DCT coefficients with value 0, or with values 0 and 1 (trying both).

For almost every cover image, the input vector at iteration contains too many wet pixels and the program fails to find the vector v. What is the way to deal with that? Do I even choose the wet DCT coefficients correctly?

1 + 3

coding-theory

steganography

kodlu

10/14/23, 12:13 PM

that reference is paywalled. you need to expand on your question

Maarten Bodewes

10/21/23, 3:24 PM

Huh? I could download without issue, though I do have research gate access. Maybe that matters or not.

Martin Benes

10/22/23, 8:02 AM

I fixed the link, no payment/access is needed now.

Score:3

Crypto

Daniel Lerch

10/14/23, 2:41 PM

It is not possible to hide a message, without modifying the wet pixels, if there are too many wet pixels. Think for example, of an image with all pixels wet: you can't modify the image without modifying some wet pixel.

The only options in these cases are to modify the wet pixels or hide less information.

Other more advanced options, such as STC, allow you to indicate a cost for each pixel. Therefore, the modification of the pixels with the lowest cost will be prioritized. In this case, the equivalent of the wet pixel is the pixel with infinite cost. But even in this case, these pixels can end up being changed if there are too many of them or if the message is too big.

You can learn more about this in the following paper, in which the amount of modified wet pixels is studied.

Minimizing Additive Distortion in Steganography Using Syndrome-Trellis Codes by T. Filler, J. Judas and J. Fridrich. [ PDF ].

+ 0

Elon Musk

I sit in a Tesla and translated this thread with Ai:

EN: Too many wet positions in Wet-Paper Codes steganography with random H

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.