Steganogan
Version: 0.1.1 Web: https://github. com/dai-lab/steganogan
concern about the security of personal data is a sign of our times. Devices, operating systems and individual programs collect information about us (hopefully anonymously), causing not a little anxiety in users. Personally, we’re not inclined to give in to such fears, and we believe that instead of fearing hidden data transmission, it’s better to take the lead and avoid it altogether. So today we’ll do some steganography and hide our super-secret text snippets in regular image files…
Steganogan is a powerful tool that applies a Torchbased deep learning algorithm to modify bitmap data. Torch itself is a quite complex machine-learning library and scientific framework, so it’s pretty useful to have Torch-powered software that’s so easy to install:
$ sudo pip3 install steganogan
Steganogan needs the above mentioned deep learning techniques to achieve so-called ‘adversarial training’. From the user’s perspective, this means that the software produces encoded images that are visually identical to their sources. Of course, it depends on the amount of details an image has, and obviously on the word count of the text to be encoded, too. The command syntax is the following:
$ steganogan encode /path/to/image “Message to hide”
Replace encode with decode to derive the text back from a previously encoded image.
The tool is very resource-intensive and that’s why it makes sense to do test runs on small images. We played with Steganogan in great depth and prepared a test image which depicted an amplified difference between the source and target images (it is possible to do this using layer blending in Gimp, for instance). The results were impressive. Suffice it to say that the steganography pattern is normally invisible and even eagle-eyed people would fail to notice any changes to a Steganogan-processed image. Let your secrets stay with you!