Segmentation, Minimum Spanning Tree and Hierarchies

Fernand Meyer, Laurent Najman · 2013

This chapter focuses on segmentation by watershed. It defines, within the framework of edge-weighted graphs, tools for segmentation: watersheds associated with image simplifications that consist of increasing floodings. Building a hierarchy of segmentations is then to force segmentations to be nested inside each other. In particular, the chapter defines some specific ultrametrics referred to as saliency maps (or ultrametric watersheds), which condense the information within the hierarchy. It focuses on some algorithms to construct saliency maps, relying on the concept of minimum spanning tree. Finally, the chapter presents a way to combine hierarchies. Controlled Vocabulary Terms image processing; image segmentation; mathematical morphology

Read the paper · More papers on PaperTik