一种改进衰减时间常数的脉冲耦合神经网络模型

An Improved Pulse Coupled Neural Network Model with Decay Time Constant

在传统脉冲耦合神经网络模型(PCNN)中,当神经元的内部活动项大于动态阈值时输出脉冲,根据输出脉冲分割出图像目标。分割弱边界的图像目标时,由于动态阈值的衰减时间常数固定不变,动态阈值的衰减速度过快或过慢于理论需要的衰减速度时,影响神经元的内部活动项与动态阈值的比较,发送错误的脉冲信号,产生误分割现象。为了解决这一问题,本文利用人眼对图像亮度的敏感性,使动态阈值的衰减速度与人眼对亮度的敏感度相一致,提出将人眼亮度权重系数作为动态阈值的衰减时间常数的系数,得到新的动态阈值。通过对含有弱边界、目标不规则、目标像素点的灰度值与背景像素点的灰度值存在重叠区域的图像进行仿真对比实验,实验结果说明本文提出的算法优于传统脉冲耦合神经网络模型。

In the classical pulse coupled neural network (PCNN) model, when the internal activity of the neu-ron is larger than the dynamic threshold, the output pulse is used to segment the image target. When segmenting the image target with weak boundary, the decay time constant of the dynamic threshold is fixed, and the decay speed of the dynamic threshold is too faster or too slower than the theoretical decay speed, which affects the comparison between the internal activity term of the neuron and the dynamic threshold, and sends the wrong pulse signal, resulting in the phenomenon of false segmentation. In order to solve this problem, this paper makes use of the sensitivity of human eyes to image brightness to make the decay rate of the dynamic threshold consistent with the sensitivity of human eyes to brightness, and proposes to take the weight coefficient of human eyes brightness as the decay time constant of the dynamic threshold, and obtains the new dynamic threshold. By comparing the images with weak boundary, irregular target and overlapping area between target pixel and background pixel, the experimental results show that the proposed algorithm is better than the classical pulse coupled neural network model.