Superpixel Based Region Segmentation Using Shallow Classifiers

This paper deals with solving the problem of semantic image segmentation in scenarios characterized by limited availability of annotated data which is common in many fields such as remote sensing, medical imaging, agriculture, etc. An approach that maps superpixels to features and uses a shallow supervised classifier is proposed. It emphasizes efficiency and context-aware segmentation, using a superpixel-based methodology. In experiments on a floodplain segmentation dataset, the algorithm shows favorable outcomes. Promising results from preliminary experiments in medical imaging also suggest potential applications in other fields.