A NEW CNN OSCILLATOR MODEL FOR PARALLEL IMAGE SEGMENTATION
Segmentation of the textured images into disjoint homogeneous regions is a very important aspect of visual perception. The texture represents properties of visualized objects; it may provide information about their structure. One of the recently developed tools used for texture segmentation is a network of synchronized oscillators. A parallel network operation is based on a "temporary correlation" theory, which attempts to explain scene recognition as performed by the human brain. This theory states that the synchronized oscillations of neuron groups attract attention if it is focused on a coherent stimulus (image object). For more than one perceived stimulus, these synchronized patterns switch in time between different neuron groups, thus forming temporal maps coding several features of the analyzed scene. Consequently, to implement this theory, a new oscillator network was proposed for image segmentation. The segmentation is obtained due to local interactions among neighboring cells. Such a network was successfully used for segmentation of the wide range of different images, including textured and biomedical ones. The network is very suitable for a hardware realization owing to its parallel structure. The realization provides a much faster image segmentation when compared to computer simulation techniques. The paper presents a new mathematical oscillator model suitable to be implemented in a CNN network chip. The model was used to design and simulate a CMOS oscillator circuit, which enables parallel network operation. The proposed oscillator model was analyzed and discussed from the point of view of its computer simulations. Furthermore, it was demonstrated that the oscillator network which implements the presented model is able to perform segmentation of the sample textured images. Oscillator circuit and block diagram of the proposed network chip were also presented and discussed.