Date of Award

Fall 2001

Document Type


Degree Name

Doctor of Philosophy in Electrical Engineering - (Ph.D.)


Electrical and Computer Engineering

First Advisor

Ali N. Akansu

Second Advisor

Nirwan Ansari

Third Advisor

Richard A. Haddad

Fourth Advisor

Mahalingam Ramkumar

Fifth Advisor

Yun Q. Shi


This dissertation is a comprehensive study of digital steganography for multimedia content protection. With the increasing development of Internet technology, protection and enforcement of multimedia property rights has become a great concern to multimedia authors and distributors. Watermarking technologies provide a possible solution for this problem.

The dissertation first briefly introduces the current watermarking schemes, including their applications in video,, image and audio. Most available embedding schemes are based on direct Spread Sequence (SS) modulation. A small value pseudo random signature sequence is embedded into the host signal and the information is extracted via correlation. The correlation detection problem is discussed at the beginning. It is concluded that the correlator is not optimum in oblivious detection. The Maximum Likelihood detector is derived and some feasible suboptimal detectors are also analyzed. Through the calculation of extraction Bit Error Rate (BER), it is revealed that the SS scheme is not very efficient due to its poor host noise suppression. The watermark domain selection problem is addressed subsequently. Some implications on hiding capacity and reliability are also studied. The last topic in SS modulation scheme is the sequence selection. The relationship between sequence bandwidth and synchronization requirement is detailed in the work. It is demonstrated that the white sequence commonly used in watermarking may not really boost watermark security.

To address the host noise suppression problem, the hidden communication is modeled as a general hypothesis testing problem and a set partitioning scheme is proposed. Simulation studies and mathematical analysis confirm that it outperforms the SS schemes in host noise suppression. The proposed scheme demonstrates improvement over the existing embedding schemes.

Data hiding in audio signals are explored next. The audio data hiding is believed a more challenging task due to the human sensitivity to audio artifacts and advanced feature of current compression techniques. The human psychoacoustic model and human music understanding are also covered in the work. Then as a typical audio perceptual compression scheme, the popular MP3 compression is visited in some length. Several schemes, amplitude modulation, phase modulation and noise substitution are presented together with some experimental results. As a case study, a music bitstream encryption scheme is proposed. In all these applications, human psychoacoustic model plays a very important role. A more advanced audio analysis model is introduced to reveal implications on music understanding. In the last part, conclusions and future research are presented.