摘要
针对脉冲耦合神经网络(pulse coupled neural network,PCNN)模型需要人工方式确定循环迭代次数,以及香农熵定义中基于对数函数存在零点处无意义的缺陷和对数运算影响处理速度等问题,提出了一种基于最小倒数交叉熵自适应生成迭代次数的PCNN图像分割算法.首先,对传统的PCNN模型进行简化,并对神经元的反馈输入函数、连接输入函数和动态阈值函数进行修正;然后,应用二维倒数交叉熵的分解算法,通过两个一维倒数交叉熵的组合获得二维倒数交叉熵;最后,采用最小倒数交叉熵准则确定PCNN网络的循环迭代次数,实现对图像的最优分割.仿真实验验证了该方法的有效性.
In view of the problem of the cyclic iteration times N of pulse-coupled neural network （PCNN） set manually, the drawback of undefined value at zero because of its definition based on logarithm and the computation speed affected by the logarithmic operation, a new method of improved PCNN image segmentation based on the criterion of minimum reciprocal cross entropy is put forward. The traditional PCNN model is simplified in this algorithm, the feeding input function, linking input function and dynamic threshold function are modified. The decomposition algorithm of two-dimensional reciprocal cross entropy is used, the two-dimensional reciprocal cross entropy is achieved by combining two one-dimensional reciprocal cross entropy. The cyclic iteration times is finally determined by the minimum reciprocal cross entropy, segmentations on various images are implemented with the proposed method. The simulation results demonstrate its validity.
出处
《扬州大学学报（自然科学版）》
CAS
北大核心
2016年第2期51-56,共6页
Journal of Yangzhou University：Natural Science Edition
基金
江苏省自然科学青年基金资助项目(BK20140266)
江苏省高校自然科学基金面上资助项目(14KJB210001)
常州大学科研启动基金项目(ZMF13020019)
江苏省高等职业院校国内高级访问学者计划资助项目(2014FX031)
关键词
图像分割
脉冲耦合神经网络
倒数交叉熵
迭代
image segmentation
pulse coupled neural network
reciprocal cross entropy
iteration
