Sensory segmentation by neural oscillators

A network of neural oscillators is proposed to segment images into figures and a background. Synchronous activity oscillations of groups of oscillators are used to label different figures in an image and to segregate them from the background. The oscillations of different groups are out of phase and show no correlations with the background activity. The oscillators are sensitive to localized input from the image and encode local feature types. The connectivity between neurons reflects the likelihood that two units belong to the same segment, i.e., all oscillators at the same location regardless of feature type and all oscillators of the same feature type regardless of their location are connected excitatorily. Global inhibition is employed as a control to achieve the segmentation task and to prevent the whole network from phase-locking. Furthermore, global inhibition enables the system to switch between synchronous and asynchronous oscillations