Segmentation is a fundamental task in image processing, enabling the decomposition of an image into meaningful regions. Among the many approaches, Mathematical Morphology provides a powerful framework for segmentation, leveraging concepts such as quasi-flat zones, watersheds, and hierarchical representations. In this talk, I will introduce key morphological tools used for (hierarchical) segmentation and discuss how Binary Partition Hierarchies (BPH) and Minimum Spanning Trees (MST) serve as essential data structures for these tasks. However, classical algorithms for BPH construction assume that the entire image fits in memory, which becomes a major bottleneck when processing large-scale data. To address this, we propose an algebraic framework alongside with efficient algorithms, allowing BPH computation in an out-of-core manner (i.e., without loading the entire image at once). Through this, we aim to make hierarchical segmentation more scalable and applicable to large images.
Localisation
Salle 5257 (ESIEE Paris)