An Oscillatory Neural Network Model of Object Selection and Segmentation in the Visual Scene

An oscillatory neural network model is presented that allows selection of a specified object from the image. The processing of the image is divided into two stages. The first stage implements contour extraction by applying traditional image processing algorithms. The result is raw contours accompanied by the noise and some spurious objects. During the second stage the searched object is segmented from the image, its boundaries are determined, and the noise is suppressed. The second stage is implemented by a two layer network of phase oscillators controlled by a central oscillator. The extraction of an object is made in terms of the temporal correlation hypothesis. The oscillators coding the selected object form an assembly with coherent activity which also runs in phase with the central oscillator. Other oscillators do not show synchrony with this assembly. The work of the model is illustrated on an example of the image used in robotics.