Rapid Anisotropic Diffusion Using Space-Variant Vision

摘要: Many computer and robot vision applications require multi-scale image analysis. Classically, this has been accomplished through the use of a linear scale-space, which is constructed by convolution visual input with Gaussian kernels varying size (scale). This shown to be equivalent solution diffusion equation on an infinite domain, as Green‘s function such system (Koenderink, 1984). Recently, much work focused variable conductance resulting in anisotropic described nonlinear partial differential (PDE). The coefficient decreasing gradient magnitude enhance edges, while some types noise (Perona Malik, 1987). Unfortunately, requires numerical integration PDE costly process when carried out uniform mesh typical image. In paper we show that complex log transformation, variants are universally used mammalian retino-cortical systems, allows integrated at exponentially enhanced rates due nonuniform spacing inherent domain. rates, coupled intrinsic compression yields speed increase between two three orders magnitude, providing means performing rapid enhancement using diffusion.