Automatic Spiking Pulse-Coupled Neural Network for Image Segmentation

Abstract

Image segmentation is the process of partitioning a digital image into multiple segments with the purpose to simplify and/or change the representation of an image. The edges identified by edge detection are often disconnected. This paper introduces a 2-D Pulse-coupled neural network for the image segmentation. A new automatic spiking pulse-coupled neural network (SPCNN) parameters setting method is presented. The dynamic of the neuron and static characteristics of image could be connected so as to ensure the parameters of SPCNN. This approach does not need the training and experiments, which is very suitable for the real-time image processing. The experiments show the outperformances with Normalized Cuts approach. From the comparison, it could be observed that given the parameter L V , if  is bigger, then the neuron in SPCNN model will be greater influenced by the neighboring neurons. That means in the space, there will be a bigger area which is formed by the neighboring neurons that can form a segmented sub-area. Additionally, e  is bigger, then the precision of the SPCNN model will be lower. That indicates the segmented sub-areas will have wider gray range in the beginning iterations of SPCNN model.