Supervisor

Suitable for

Abstract

Bag gain is something that happens when a sender wishes to use steganography to spread a secret message across a number of covers: the set of objects sent, some of which contain the hidden payload, is called a bag. Theory predicts that the size of the secret that can be undetectably transmitted should scale with the square root of the size of the bag, but in practice researchers have observed that it grows faster. This is attributed to being able to select only the "best" covers in the bag, where "best" means those in which the presence hidden data is hardest to detect (for example, noisy images).

This project, which is experimental in nature, aims to replicate and extend these observations. The student will need to use off-the-shelf implementations of simple steganography in images, with an image library (supplied by the supervisor), implementing a steganography detector by combining off-the-shelf implementations, which will need an ability to train CNNs (probably using pytorch, but other packages may also be suitable). Experiments will determine how the detectability of hidden data depends its size, the size of the bag, and the method used to spread the message into the bag. The results will then be analyzed.