A new concept that uses a pyramidal approach in image segmentation is proposed. The method introduces local visual pyramids in reply to inherent limitations of the classical pyramidal structure (e.g. concerning small or elongated objects). the aim of local visual pyramids is to simulate the human vision in its attention focusing processes through an individual and a convtextual analysis. Actually, a local visual pyramid is a hierarchy of fine to coarse resolution versions of an image where the resolution decrease twofold between consecutive levels. The kernel of such a pyramid is defined using the distribution of human visual acuity i.e. the kernel weights visual information roughly by an inverse function of the visual angle. Such an approach allows to improve the precision and the robustness of pyramid-based segmentation processes.
展开▼