Texture Segmentation by Local Bi-Orthogonal Decomposition

We investigate the ability of a local bi-orthogonal decomposition to build texture segmentation of images. Using the structures associated to the local decomposition of the image independent row and columns we perform a segmentation, where the regions are defined by the property of having a smooth variation of the corresponding entropy. Examples are choosen in texture made and also in real life images. The size of the local analysis is also determined by the properties of the (global) bi-orthogonal decomposition.