Jordan features for texture segmentation

This paper deals with the application to texture segmentation of two concepts, namely the Jordan Decomposition Theorem, which says that any bounded variation function can be expressed as the difference of two non-decreasing functions, and the Peano-scan, which is a 1D traversal of square domains. In this framework, a family of so-called Jordan decomposition for images are proposed. Each decomposition provides a representation of the image as an ordered pair of non-decreasing functions. These representations are used to define a set of features, called Jordan features, which give informations on the local gray-level distribution. Applications to texture segmentation are presented showing promising results