<p>All known methods of lossless or reversible data embedding that exist today suffer from two major disadvantages: 1) The embedded image suffers from distortion, however small it may be by the very process of embedding and 2) The requirement of a special parser (decoder), which is necessary for the client to remove the embedded data in order to obtain the original image (lossless). We propose a novel lossless data embedding method where both these disadvantages are circumvented. Zero-distortion lossless data embedding (ZeroD-LDE) claims 'zero-distortion' of the embedded image for all viewing purposes and further maintaining that clients without any specialized parser can still recover the original image losslessly but would not have direct access to the embedded data. The fact that not all gray levels are used by most images is exploited to embed data by selective lossless compression of run-lengths of zeros (or any compressible pattern). Contiguous runs of zeros are changed such that the leading zero is made equal to the maximum original intensity plus the run-length and the succeeding zeros are converted to the embedded data (plus maximum original intensity) thus achieving extremely high embedding capacities. This way, the histograms of the host-data and the embedded data do not overlap and hence we can obtain zero-distortion by using the window-level setting of standard DICOM viewers. The embedded image is thus not only DICOM compatible but also zero-distortion visually and requires no clinical validation.</p>
N. Nagaraj and Mullick, R., “Zero-distortion Lossless Data Embedding”, in Medical Imaging 2004, 2004.