James Hahn, a Computer Science major at SCI, recently completed a THINK Fellowship project with Pitt’s Honors College. His thesis, “Hiding in Plain Sight: Chicken Scratch and Applications to Steganography,” looks into hiding secret information in normal data. Steganography is an emerging field in computer science, which has only recently made significant strides.

Specifically, James took normal-looking images, trained a neural network, and encrypted data into the image. This task is challenging because it is two-fold: it must pass human visual inspection and machine inspection. If a human looks at the output image and cannot directly observe the hidden information/data, then the autoencoder did a good job. The second task, machine inspection, is trying to prevent a computer, or machine-learning algorithm, from detecting patterns in the data to decrypt, uncover, and reconstruct the secret information.

James is particularly looking at this from the perspective of handwriting on a white background. If an image is saturated with color, the visual inspection task will be easier since changing a shade of blue is not noticeable to the human eye. If you change a white to shade of grey, this is more noticeable. He is approaching this problem with two new ideas. First, he is encrypting the secret data. Then he is trying to hide secret information in the actual stroke data of a person’s handwriting.

James spoke on the different uses on steganography saying, “…if a movie studio pre-releases a film with a unique watermark for every person they give it to, and then a copy is leaked, they can use that copy of the film to trace it back to the exact perpetrator.” Another example of steganography is an artist who takes a photograph and places it on the internet. They can attach a completely hidden watermark to the photograph and use it in copyright cases in court.