We propose a novel variational segmentation method that considers shape and size adaptive patches to characterize the regions of the image that share the same local structure up to an affine distortion. They are used in an L1-norm energy model that uses fuzzy membership functions and its total variation. The output of the method is a partition of the image in regions together with a representative patch of the texture of each region.

This is poster number 28 in Poster Session

Authors:

Maria Oliver (Universitat Pompeu Fabra)

Gloria Haro (Universitat Pompeu Fabra)

Vadim Fedorov (Universitat Pompeu Fabra)

Coloma Ballester (Universitat Pompeu Fabra)

Keywords:

image segmentation, l1-energy, patch-based methods, variational methods