A hierarchical image segmentation oscillator network based on shared contextual synchronization

A hierarchical oscillator network is designed and inspired by the context vision idea. The proposed network is a three-layer neuron structure, that is “perceptual layer”, “component layer” and “object layer”. The strategy is to first divide the image into tiles with overlapping parts, use parallel processing method to finish first layer basic segmentation for each tile independently, and use one neuron for each recognized component; then use the context characteristics to build connections merging the components into objects; finally look up the indexes of objects back to the pixels and synthesize the result picture use an average grey scale for each object. The algorithm has been successfully implemented on a “Q” shape pattern to examine the validity and further on a sample MRI image to prove the processing effect. The designed algorithm utilizes the oscillator synchronization imitating the vision mechanism in human brain to deal with image processing problem. The idea is promising and initially proved to be plausible.