Self-Healing Images

Principal Investigator: Hany Abdel-Khalik

The fragility of highly sensitive images has become a subject of recent concern to monitor and recover information in response to data corruption. The so-called self-healing image refers to the capability to restore lost information from a corrupted image, i.e., correcting pixelation or compression-induced imperfections. Furthermore, these methods serve as a self-authentication method in which the image can be confirmed as genuine by comparing top-level pixel information to embedded details; if there is a large discrepancy in the two, then it is clear that the images have been tampered with. The main purpose of this project is to construct a novel information processing algorithm that is applicable to multimodal data such as images and timeseries, wherein the true data is reinforced by self-healing capabilities, i.e., the ability to recover lost or damaged information.

The difficulty with typical methods is the ability to convey image clarity inversely decreases with the information embedded; one such method is adding information-carrying noise to the image, and many others utilize water-marking techniques. While the information potentially conveyed is very high, image clarity is quickly sacrificed in this procedure. The method proposed in this project allows for unrelated information-carrying components, e.g., timeseries-based information, another image’s information, etc. to be embedded without compromising top-level behavior or visual acuity .

This method distinctly deviates from watermarking, which is overtly added to the data, and steganography, which can be discovered and reverse engineered, in that the proposed method is cryptographically secure by virtue of a randomized protocol that embeds information into the null space of a given image.

Personnel

Students: Arvind Sundaram Dylan Adal

Keywords: intrusion detection, self-healing data, subtle anomaly. information recovery