Best Paper Prize for Devis Tuia et al. at PCV-ECCV in Zürich !
© 2014 EPFL
Devis Tuia from LaSIG, and co-authors, were awarded an honorable mention Best Paper Prize at the ISPRS mid-term Symposium "Photogrammetric Computer Vision" (PCV) held in conjuction with the European Conference on Computer Vision (ECCV) in Zurich, Switzerland.
The paper awarded, entitled "A group-lasso active set strategy for multiclass hyperspectral image classification", deals with landuse classification from hyperspectral imagery. More specifically, it proposed a new efficient and automatic strategy for the generation and selection of contextual filters improving the quality of the classification. Based on a fast and sparse multiclass logistic classifier, the model can process large hypersepctral images and discover the relevant filters to distinguish several types of land use, even in presence of similar materials. The work was performed in collaboration with research teams at the University of Nice Sophia Antipolis and with the University of South Brittany.