Digital Data-Hiding Pulls a Reverse
Daniel C. McCarthy
The ability to easily modify and manipulate digital imagery has been beneficial to applications in the entertainment and consumer markets. However, this mutability poses problems in military, legal, medical and other fields where the accuracy and veracity of an image are vital. Digital watermarks or "data-hiding" algorithms can help guarantee an image's validity, but even if removed, they often alter details in the image, preventing it from being enlarged or enhanced for further detail.
Which 512 x 512 gray-scale image includes more data? Although the image on the right carries 3400 additional bytes, the extra data is imperceptible. Unlike conventional data-hiding techniques, in this approach, the embedded data can be extracted without loss or alteration to the original image. Courtesy of the University of Rochester.
This may not matter for applications in which validity or security matter more than fidelity to the original. But for those where fidelity does matter, a reversible data-hiding technique has been developed at the University of Rochester in New York and at Xerox Corp. in Webster, N.Y.
Like other methods, reversible data-hiding can help detect whether anyone has altered the image. Conventional watermarking algorithms embed information by modulating a pseudo noise pattern and adding the modulated pattern to the image. A correlation-based detector identifies and extracts the embedded information but does not preserve the original state of the image.
The new technique extracts imperceptible image features -- such as the least significant bit of each pixel value -- and compresses them without loss. Then it modifies the features to embed the compressed data, along with additional payload information, into the image. Extraction reverses the steps, obtaining the embedded information, decompressing the original features and substituting them for the digitally altered features.
Potential applications include insertion of patient information into a computed tomography scan or x-ray. Like other data-embedding algorithms, this method bonds data to the image, preventing errors when a file is misplaced or its file name mistakenly modified, said Mehmet Celik, a research assistant at the university. He explained that other techniques have introduced irreversible distortions and caused loss of information.
"Although these distortions are mostly imperceptible or tolerable, the information loss may have diagnostic consequences and possible liability implications," he said. "Our algorithm guarantees a mathematically lossless system where the original image is reproduced exactly after extraction of the embedded information. When used together with digital signatures, it provides strong assurances about the integrity of the image."
The researchers are working on enabling their algorithm to embed more data and on developing a reversible algorithm to detect the location of any tampering.
MORE FROM PHOTONICS MEDIA