Partition filters: a new class of morphological operators for segmenting textured images

Morphological partition filters are introduced in the context of textured-image segmentation. These filters are characterized by a controllable selectivity of action with respect to both directivity and spatial frequency. They are thus able to homogenize regions whose structure corresponds to their characteristics, while leaving others untouched. Automatic adjustment of their structuring elements is obtained through computation of the image covariograms during the various stages of the procedure. The proposed segmentation technique was tested on images made up of natural textures. The procedure appears to perform well in cases of highly structured textures