Using local pyramids for more robust object delineation

Multiresolution techniques are often used for image segmentation, emulating the human visual process. More precisely, image pyramids are especially well adapted for multiresolution representation. A new concept is proposed that uses a pyramid approach in image segmentation, by simulating the human vision in its attention focusing process. The method introduces local pyramids in reply to inherent limitations of the classical structure, especially the case of elongated or countless objects. To show the robustness of such a new technique, experimental results and comparisons with classical methods are given