Coding and data hiding for multimedia

Fast advances in computer hardware and networks have facilitated rapid growth in digital multimedia data. Sophisticated compression schemes are needed to reduce the growing demands of storage space and communication bandwidth of multimedia data without noticeable degradation of the data. On the other hand, fast growth of digital multimedia data has also opened up the possibility to hide information within multimedia files without any interference to normal applications of these files. It is also possible for two end users to communicate with each other without any visible communication channels.; This thesis addresses two key issues in multimedia applications. The first issue is to design efficient, high perceptual quality compression algorithms to meet the demand in multimedia applications. We introduce a novel image compression algorithm which employs adaptive signal representations and exploits directly human visual masking to maximize reduction of both redundancy and, irrelevancy in an image. In addition, we also introduce a new arithmetic coding procedure, and a new tree coding algorithm, together with theoretical performance analysis. The strengths of all our coding algorithms are demonstrated with experimental results.; The second issue this thesis addresses is to hide information into digital media such as image, audio or video signals. We introduce two robust data hiding algorithms for multimedia data hiding. Both algorithms exploit directly the masking phenomena in the human visual and auditory systems to maximize energy of the embedded data without incurring any perceptual degradation to the host media. Data is embedded in such a way that it is perceptually and statistically undetectable. One of the algorithms is truly secure , i.e., without the key, it is computationally impossible to extract the hidden data. We illustrate through experimental results the fidelity and robustness of hidden data to lossy operations such as compression, additive noises, lowpass filtering and decimation.; To help design perception based coding and hiding algorithms, we introduce two masking models for visual data. One is a frequency masking model that is developed from experimental data in psychophysics. The other is a spatial masking model that comes from modifications to an existing threshold vision model.