Analog Implementation of Segmentation Algorithm Using Oscillator Network

We propose a bio-inspired visual processing system for pattern segmentation which parallelizes the information and shows high-speed performance and flexible features. Locally Excitatory Globally Inhibitory Network (LEGION) is one of the bio-inspired visual processing systems. The network is composed of an array of oscillators and Global Inhibitor. In the network, a global element had functions to receive input from the entire network and feedback. We propose a network structure in which each oscillator module shares the function and the global element can be regarded as a data field. As a result, we successfully enhance the features as a parallel and distributed system. We also propose a new oscillator model which has three modes; charge, discharge and stop. The period of oscillation is controlled by means of switching these modes and it enables the network to easily achieves both synchronization within oscillators which belong to the same object and desynchronization between different objects. We indicate the feature that the period of oscillation is proportional to the number of objects. We verify these features by simulations and experiments with real implementation of analog circuit and show the result of pattern segmentation with it.