一种参数自适应的简化PCNN图像分割方法

摘要

In order to extend the pulse coupled neural network (PCNN) to image segmentation, this paper improves the original PCNN model and simplifies its parameters, and then presents a method for adaptive parameter settings. The proposed method adopts the relationship between the neural threshold and the region characteristic of pulse output. Meanwhile, an optimal method is introduced to find the optimal linking coeffcient value. As a result, our model can iteratively segments the targets from the background. Experimental results on synthetic and real infrared images demon-strate that the proposed method has better adaptability to the complicated images when compared with the widely used threshold and two state-of-the-art PCNN models, and shows good robustness for noisy images.%为了进一步延伸脉冲耦合神经网络(Pulse coupled neural network, PCNN)在图像分割中的应用,本文对PCNN模型作了简化和改进,并探讨和分析了参数的设置方法。首先利用阈值和脉冲输出所对应的区域均值之间的关系,提出了一种优化连接系数的方法,使得模型最终以迭代的方式得到分割结果。在仿真和真实红外图像上实验结果表明,文中方法能取得较优的分割效果,且相比于常用的阈值方法以及较新的PCNN 方法,文中的简化模型对噪声及复杂图像具有更好的适应性和鲁棒性。