The current work provides a new data-embedding infrastructure based on fractal modulation. The embedding problem is tackled from a communications point of view. The data to be embedded becomes the signal to be transmitted through a watermark channel. The channel could be the image itself or some manipulation of the image. The image self noise and noise due to attacks are the two sources of noise in this paradigm. At the receiver, the image self noise has to be suppressed, while noise due to the attacks may sometimes be predicted and inverted. The concepts of fractal modulation and deterministic self-similar signals are extended to 2-dimensional images. These novel techniques are used to build a deterministic bi-homogenous watermark signal that embodies the binary data to be embedded. The binary data to be embedded, is repeated and scaled with different amplitudes at each level and is used as the wavelet decomposition pyramid. The binary data is appended with special marking data, which is used during demodulation, to identify and correct unreliable or distorted blocks of wavelet coefficients. This specially constructed pyramid is inverted using the inverse discrete wavelet transform to obtain the self-similar watermark signal. In the data embedding stage, the well-established linear additive technique is used to add the watermark signal to the cover image, to generate the watermarked (stego) image. Data extraction from a potential stego image is done using diversity combining. Neither the original image nor the original binary sequence (or watermark signal) is required during the extraction. A prediction of the original image is obtained using a cross-shaped window and is used to suppress the image self noise in the potential stego image. The resulting signal is then decomposed using the discrete wavelet transform. The number of levels and the wavelet used are the same as those used in the watermark signal generation stage. A thresholding process similar to wavelet de-noising is used to identify whether a particular coefficient is reliable or not. A decision is made as to whether a block is reliable or not based on the marking data present in each block and sometimes corrections are applied to the blocks. Finally the selected blocks are combined based on the diversity combining strategy to extract the embedded binary data.

Library of Congress Subject Headings

Digital communications; Digital watermarking; Image processing--Digital techniques; Digital media; Digital modulation

Publication Date


Document Type


Student Type


Degree Name

Electrical Engineering (MS)

Department, Program, or Center

Electrical Engineering (KGCOE)


Raghuveer Rao

Advisor/Committee Member

Ferat Sahin

Advisor/Committee Member

Daniel Phillips


Physical copy available from RIT's Wallace Library at TK5103.7 .A76 2004


RIT – Main Campus