SCENE SEGMENTATION USING SOLELY EXCITATORY OSCILLATOR NETWORKS (SEON)

摘要

A solely excitatory oscillator network (SEON) is proposed for scene segmentation. SEON achieves synchronization and desynchronization between neural oscillators rapidly in an oscillator network. The synchronization speed is independent of the number of oscillators, enabling rapid synchronization in a very large network. SEON adopts a segmentation method that produces reliable segmentation results in a highly parallel manner. The segmentation speed will not decrease in a very large network, allowing SEON to utilize its invariant synchronization speed properties. Using a gallery of 1200 images, our model shows an average segmentation rate of over 98%. It produces continuous boundaries and is very efficient in the detection of vague boundaries. Compared with other contemporary segmentation methods, SEON provides promising results in both segmentation power and speed. The improvement of speed becomes more significant for large images.