Segmentation of Multisource Image by Multiview Learning

Image segmentation is a critical technique for remote sensing image analysis and interpretation. It is the basis of image understanding, such as region-based change detection for maps updating, target recognition, and so on. Existing literatures [1, 2] have proved that, by combining multiple data sources, e.g., synthetic aperture radar (SAR) and optical imagery, the overall classification accuracy will significantly increase compared to the quality of a single-source classifier. Multi-view learning approaches [3, 4, 5] form a class of semi-supervised learning techniques using multiple views to effectively learn from partially labeled data. Although there are a variety of multi-view learning algorithms, they are all founded on the common principle: training the classifiers for the views by maximizing the agreement on their predictions for the unlabeled data 2 ( ) ( )