Abstract: Aiming at the several problems that parameters are numerous and difficult to select automatically and the ending condition of the iteration is not well determined in the traditional PCNN model, a method of blood vessels segmentation in fundus image based on gray iteration threshold PCNN is proposed. The traditional PCNN model is simplified and link input is improved into the sum of neuronal excitatory link input and inhibitory burst link input of neuronal neighborhood. Meanwhile, the exponentially decaying threshold is improved to the grayscale?iteration?threshold. When the method is applied to image segmentation, parameter or specific criteria doesn’t need to be set to determine the best number of iterations since the segmentation can be completed by one time of PCNN iterating process. The experiments implemented on the DRIVE fundus image database indicate that the method is superior to traditional PCNN method in subjective visual effect and objective segmentation performance and operation time.

Key words: vascular segmentation, pulse coupled neural network, link input, gray iteration threshold, segmentation performance

摘要： 针对传统的脉冲耦合神经网络（Pulse Coupled Neural Network，PCNN）模型中参数众多且不易自动选取，迭代次数结束条件不好确定的问题，提出了一种基于灰度迭代阈值脉冲耦合神经网络的眼底图像血管分割方法。该方法简化了传统PCNN模型，将其单一的神经元兴奋性链接输入改进为神经元兴奋性与邻域抑制性链接输入之和；同时将其随时间指数衰减的阈值改进为图像的灰度迭代阈值，分割图像时无需人工设置参数，无需特定准则确定最佳迭代次数，一次迭代完成分割。对DRIVE眼底图像库的实验结果表明，该方法在主观视觉效果及客观分割性能和运算耗时上均明显优于传统PCNN方法。

关键词: 血管分割, 脉冲耦合神经网络, 链接输入, 灰度迭代阈值, 分割性能