Segmentation of natural images based on multiresolution pyramids linking of the parameters of an autoregressive rotation invariant model. Application to leather defects detection

Random field (RF) parameters can be subjects of a probability distribution useful for image segmentation. A multiresolution methodology, based on the node linking of a pyramid that takes into account some kind of a Gaussian distribution for node values and its geometry is applied to the RF parameters to segment the natural image segmented and enhance some of the immersed objects. Before applying the above described system to a case study (calf leather), a simulation for the multiresolution methodology is applied to a generated RF, having RF objects with different means