Remote sensing image segmentation using neural oscillator network of visual cortex

Locally excitatory globally inhibitory oscillator network (LEGION), a neural oscillator network based on biologically framework, is applied to remote sensing image segmentation. For remote sensing images with complicate environment, a simplified LEGION algorithm with improved anisotropic diffusion method is presented to overcome sensitiveness. In multi-spectral images all the pixel values of the same location is considered as a vector. The connection weight between two oscillators is determined by a Mahalanobis distance metric between the two corresponding vectors. Experiments on real panchromatic image show that the proposed method can effectively produce more accurate segmentation than traditional algorithms. And the extending LEGION image segmentation algorithm can also be applied to process multi-spectral images segmentation.