Image segmentation based on modified pulse-coupled neural networks

Abstract

Based on the property of Human vision system (HVS) that human eye's sensitivity to an image varies with different information regions of the image, Pulse-coupled neural network (PCNN) model is modified for image segmentation. The modified PCNN stimulated by an input image has pulse output with multiple pulse values rather than only two in the conventional PCNN, according to the local information rate of the input image. This results in image segmentation according to local information rate delivered from the image by the modified PCNN. Experiments with the modified PCNN on image segmentation and image compression on the segmented images with the principle that the lower information rate is, the higher compression rate is applied, show much better performance in compression rate compared with that on the segmented images with the conventional PCNN.