A Picture is Worth a Thousand Words, Literally
AI-powered steganography hiding data in social network images beyond human visual perception.
Summary
Digital Media Communication
Images, videos, and digital media provide convenient communication through social networks. Broadcasting capability and information richness enable both legitimate and illicit uses.
Image Steganography
JPEG files can disguise foreign data through minor pixel and metadata alterations. Covert messages in billions of daily posted images might be possible, but scale automation faces technical challenges.
Neural Network Learning
Convolutional neural networks learn to reverse-engineer transformations applied by social networks. The network optimizes hidden data capacity by identifying pixels and metadata least modified during platform processing.
Data Acquisition and Results
Social networks provide free bulk upload/download for training data. Experiments show hidden data transmission through social images with high fidelity, demonstrating AI can hide data in plain sight at scale beyond human detection despite third-party manipulation and compression.
Related talks
AI Village·22 min
Sven Cattell
Deep learning approaches to security threats and vulnerabilities.
AI Village·27 min
Matt
Neural network approaches to cybersecurity challenges.
AI Village·25 min
Joseph Clements
Designing hardware trojan exploits embedded directly into neural network accelerator silicon and hardware.