Application Of Oscillator Neural Network In Image Segmentation

Coupled systems have come to be a hot research spot of non-linear dynamics.Lots of experiments confirm that there is a phenomenon of synchronization in coupledsystems. Synchronization ofen occurs in coupled neurons which is ofen observed inmany physiological experiments. The synchronization of coupled neurons has been acrucial problem in coupled systems, particularly for the oscillation synchronization.According to the temporal correlation theory, the neuron clusters process theimage segmentation, edge detection, associative memory, pattern recognition andother problems based on synchronous oscillation. So many people take advantage ofneurons which there exists a limit circle in a single neuron to form a local couplingneural network for solving visual information processing. Deliang Wang, DavidTerman and others applied the LEGION model with only one limit circle in imagesegmentation, which simulates the visual neurons. While the neurons oscillate in thefrequency of γ(30Hz~80Hz), so there should be multiple limit circles in the neuralmodel, so that they can oscilltate in different frequency. So we should use the neuronswith many limit circles to analog visual neurons.Based on the oscillation synchronization, the research mostly of this topic is asfollowing:(1) we study a kind of special equation with two stable limit circles andone unstable limit circle. Also we add coupled terms in this kind of equation andapply it in image segmentation.(2) we improve original Kuramoto model, whichphase coupling is changed to frequency coupling and globally coupling is changed tobe local coupling, we apply it in image segmentation.(3)We couple the improvedKuramoto model with the center neuron to simulate attention selection process.