Semantic rich ICM algorithm for VHR satellite images segmentation

Jérémie Sublime (1) , Andrès Troya-Galvis (2) , Younès Bennani (3) , Antoine Cornuéjols (1) , Pierre Gancarski (2)

1 MIA-Paris - Mathématiques et Informatique Appliquées
2 ICube - Laboratoire des sciences de l'ingénieur, de l'informatique et de l'imagerie
3 LIPN - Laboratoire d'Informatique de Paris-Nord

In this article we show some applications of a MRF-based segmentation algorithm applied to real data extracted from a very high resolution image. This algorithm has specific features that enable the extraction of semantic information on the clusters in the form of affinity and geographic position properties. The results of the experiments conducted on this data set are interesting both in terms of clustering quality when using common unsupervised learning quality indexes, but also when compared to a ground-truth based on expert maps.

Domains: Life Sciences [q-bio]

Jérémie Sublime, Andrès Troya-Galvis, Younès Bennani, Antoine Cornuéjols, Pierre Gancarski. Semantic rich ICM algorithm for VHR satellite images segmentation. 14th IAPR International Conference on Machine Vision Applications (MVA), May 2015, Tokyo, Japan. pp.15292961, ⟨10.1109/MVA.2015.7153129⟩. ⟨hal-01585544⟩