| Image segmentation plays an essential role in image processing and often serves as the basis for further goal-specific tasks. With the increase in the speed and amount of data generated and distributed nowadays, the need for unsupervised image segmentation has also become more important, either as a stand-alone module or as a computerized tool to assist human operators. In this dissertation, we propose a data investigation model that both prescribes the process and outlines the components of unsupervised segmentation. Subsequently we focus the application of the model to Synthetic Aperture Radar (SAR) sea ice imagery whose analyses are important in global climate monitoring and polar navigation. That this imagery bears dynamic, complicated, and varying image characteristics due to seasonal and geographical differences also provides us a practical and challenging domain to employ and test our model and applications.; We show in this dissertation that we have conducted extensive and intensive studies in the areas of unsupervised segmentation, which have led us into much work in image processing, machine learning, data mining, and remote sensing. As a result, we contribute to these various disciplines with our creation of algorithms and concepts as well as insightful findings.; Finally, we demonstrate that our model is general and applicable to different domains, and, through our presentation and discussion of results, that we have accomplished the task of unsupervised segmentation in SAR sea ice image analysis. |
