Abstract
[Truncated] This thesis addresses image and video sequence segmentation, with emphasis placed on three topics: still image low-level segmentation, video object plane (VOP) segmentation, and region-based enhancement of images degraded by coding artifacts.
Classical Bayesian low-level segmentation algorithms suffer from several shortcomings. These are the necessity of an initial estimate and the dependence on the input parameter K, denoting the number of labels to be used. To avoid these weaknesses, a new approach is proposed. The major contribution in this approach is the separation of the initialization step from the actual labeling process. This was previously proposed for morphological segmentation, but has not yet been applied to Bayesian methods. The described segmentation algorithm can therefore be seen as a combination of the advantages of Bayesian and morphological techniques.
Classical Bayesian low-level segmentation algorithms suffer from several shortcomings. These are the necessity of an initial estimate and the dependence on the input parameter K, denoting the number of labels to be used. To avoid these weaknesses, a new approach is proposed. The major contribution in this approach is the separation of the initialization step from the actual labeling process. This was previously proposed for morphological segmentation, but has not yet been applied to Bayesian methods. The described segmentation algorithm can therefore be seen as a combination of the advantages of Bayesian and morphological techniques.
| Original language | English |
|---|---|
| Qualification | Doctor of Philosophy |
| Awarding Institution |
|
| DOIs | |
| Publication status | Unpublished - 1998 |
Take-down notice
- This thesis has been made available in the UWA Profiles and Research Repository as part of a UWA Library project to digitise and make available theses completed before 2003. If you are the author of this thesis and would like it removed from the UWA Profiles and Research Repository, please contact [email protected]