Pulse coupled neural network for image processing

The results of recent studies of the visual cortices of cats and monkeys have led to the development of a new class of artificial neuron models. These models resemble biological neurons more closely than the commonly used artificial neurons. Eckhorn and his co-workers have developed one such neuron, referred to as Eckhorn's neuron. They have demonstrated that the recurrent networks of Eckhorn's neurons are capable of duplicating some of the neuro-physiological phenomena observed in cat's visual cortex.;Simulation results indicate that the performance of the PCNN based smoothing technique is better than the performance of the neighborhood averaging and median filtering techniques. The PCNN based segmentation outperforms some of the commonly used segmentation methods--intensity thresholding, optimal thresholding and region growing techniques. The PCNN approach yields results that are comparable to the results obtained by the probabilistic relaxation technique, but without the restrictions associated with the relaxation technique. The iterative knowledge based object detection system developed in this dissertation shows that the PCNNs being powerful and flexible have potential in real-time image processing systems.;In this dissertation Eckhorn's neuron model has been modified so that the resulting neuron, referred to as the pulse coupled neuron (PCN), becomes more suitable for image processing applications than his original model. It is shown that a single layered laterally connected pulse coupled neural network (PCNN) is capable of smoothing and segmenting digital images. It is also shown that the PCNN can be utilized for detecting objects in digital images.