Connected AI: Merging Mathematical Morphology and Machine Learning

Connected morphological filters allow efficient filtering, segmentation and classification of images of huge sizes, using mathematical models of perceptual grouping based on connectivity. Using them as adaptive feature extractors for machine-learning methods gives the latter access to non-local information, in ways not possible with other techniques. Furthermore, the features extracted can be scale, rotation, and translation invariant, removing the need for common data augmentation methods. These abilities, plus the development of efficient parallel and distributed algorithms allows application in fields such as remote sensing for precision agriculture using drones, and disaster relief using satellite or aerial images, but also in astronomy, microscopy and medical imaging. This poster gives an overview of the advances so far.