An evaluation of the number of bits that can be hidden within an image by means of frequency-domain watermarking is given. Watermarking is assumed to consist in the modification of a set of full-frame DCT (DFT) coefficients. The amount of modification each coefficient undergoes is proportional to the magnitude of the coefficient itself, so that an additive- multiplicative embedding rule results. The watermark-channel is modeled by letting the watermark be the signal and the image coefficients the noise introduced by the channel. To derive the capacity of each coefficient, the input (i.e. the watermark) and the output (i.e. the watermarked coefficients) of the channel are quantized, thus leading to a discrete- input, discrete-output model. Capacity is evaluated by computing the channel transition matrix and by maximizing the mutual input/output information. Though the results we obtained do not take into account attacks, they represent a useful indication about the amount of information that can be hidden within a single image.

Date Published: 9 April 1999
PDF: 12 pages
Proc. SPIE 3657, Security and Watermarking of Multimedia Contents, (9 April 1999); doi: 10.1117/12.344694

Published in SPIE Proceedings Vol. 3657:
Security and Watermarking of Multimedia Contents
Ping Wah Wong; Edward J. Delp III, Editor(s)