Semagrams are a subset of steganography. When a message is transmitted in a non-textual format, (i.e., in the visual content of an image), it is referred to as a semagram. While semagrams are relatively easy to create (as shown in published papers covering hiding techniques), detecting a hidden message in or embedded as an image-based semagram is a greater magnitude of difficultly than typical digital steganography. US Patents issued based on semagram technology show that this feature has been exploited in the copyright/watermarking world to increase protection. In a semagram, the image is the message and they work well for simple messages and dead drops. Attacks on semagrams are primarily visual examinations of artifacts. In the counter-espionage world, the rule of the thumb is that there is always a message hidden in an image or graphic, it is simply up to the steganalyst to find it. In short, detecting semagrams is a matter of recognizing patterns of patterns that represent a hidden message within an image. This presentation provides a brief summary of the technology underlying semagrams, present a short non-technical discussion of the technology used in the attack on semagrams, followed by a discussion on current work and planned future implementations of the proven semagram detection ANN. It will focus on extending the ANN to other domains (e. g., non-visual spectrums, multi/cross spectrum correlation, scene identification, image classification) and efforts to improve the processing speed and throughput via parallel/distributed methods.
