Oscillatory Network for Synchronization-Based Adaptive Image Segmentation

Oscillatory network model with controllable oscillator dynamics and self-organized dynamical coupling has been created for synchronization-based image processing. The model was previously obtained via reduction from a biologically motivated oscillatory model of the brain primary visual cortex. The reduced network model performance consists in network relaxation into the state of synchronization. The set of internally synchronized but mutually desynchronized network ensembles (clusters), arising at final synchronization state, corresponds to full set of image fragments. New model developments, presented in the paper, include: a) the advanced version of single oscillator dynamics, admitting introduction of arbitrary continuous dependence of oscillator limit cycle size on pixel brightness; b) new principle of network coupling, permitting to raise image segmentation accuracy and to control network noise reduction. A capability of selective image segmentation (extraction of image fragment subset of a priori prescribed brightness levels) is also inherent to current model version.