Texture segmentation considering multiband, multiresolution and affine invariant roughness

Texture segmentation is a basic task in the process of satellite image analysis. However, texture characterization is not an easy task specially when more than one wavelength band need to be considered. Several difficulties are present, as the basic texture definition in different resolution and image scales. Consequently, some approaches are essentially empiric and should be adjusted to different needs. We present a new method for combining multiband information for texture segmentation. It is based on an extension of fractal dimension analyze of texture for multi channel and is rotational invariant This method allow texture classification of thematic maps made from combination of N wavelength bands. The method was validated comparing its results with others implementations using mosaic of natural textures and real satellite images.