Complete lattice learning for multivariate mathematical morphology

摘要: Display Omitted Extends mathematical morphology to multivariate vectors.Proposes an efficient strategy for complete lattice learning.Requires no prior assumption on background/foreground.Can integrate supervised information.Enables perform patch-based morphological operations. The generalization of vector spaces is addressed in this paper. proposed approach fully unsupervised and consists learning a from image as nonlinear bijective mapping, interpreted the form learned rank transformation together with ordering vectors. This vectors relies three steps: dictionary learning, manifold out sample extension. In addition providing way construct vectorial ordering, can become by integration pairwise constraints. performance illustrated color processing examples.