An Iterative Thresholding Segmentation Model Using a Modified Pulse Coupled Neural Network

This paper presents a novel iterative thresholding segmentation method based on a modified pulse coupled neural network (PCNN) for partitioning pixels carefully into a corresponding cluster. In the modified model, we initially simplify the two inputs of the original PCNN, and then construct a global neural threshold instead of the original threshold under the specified condition that the neuron will keep on firing once it begins. This threshold is shown to be the cluster center of a region in which corresponding neurons fire, and which can be adaptively updated as soon as neighboring neurons are captured. We then propose a method for automatically adjusting the linking coefficient based on the minimum weighted center distance function. Through iteration, the threshold can be made to converge at the possible real center of object region, thus ensuring that the final result will be obtained automatically. Finally, experiments on several infrared images demonstrate the efficiency of our proposed model. Moreover, based on comparisons with two efficient thresholding methods, a number of PCNN-based models show that our proposed model can segment images with high performance.