Modeling location information for wavelet-based image coding.

다국어 초록 (Multilingual Abstract)

This thesis looks at the problem of wavelet image coding with special emphasis given to edges. Nonlinear representations and models are constructed in wavelet domain so as to provide accurate and effective processing of edge information for natural images.
Adaptive wavelet coders point towards the importance of locating significant information in wavelet domain. This location information is inherently linked with the properties of edge contours. Therefore, a better understanding of edge behavior is crucial for improving model adaptation in wavelet subbands. In the first part of this work, a simple analytical model is constructed to parameterize the relationships between edges and wavelet structures. The model leads to an adaptive coding strategy that highlights both the potential and difficulty of edge-based processing of image information.
The main challenge in modeling edge information is to capture the variety of legitimate edge behavior in real-life images. A realistic location representation is needed in order to build accurate models. The rest of the thesis focuses on the development of a non linear framework that has the flexibility to deal with the interactions of value and location information. The spherical coding algorithm is introduced as a successful application of location-based descriptions to coding wavelet subbands.
Finally, the generic location description of the spherical coder is modified to represent edge contours exclusively, leading to single quadrant complex signal representation. Magnitude and phase components are shown to draw an accurate picture for characterizing underlying edge contours. A nonredundant representation is made available through the design of perfect reconstruction complex filterbanks. The thesis is concluded with a detailed discussion of various issues related to modeling and coding this representation.