一种复杂背景环境下的改进型PCNN图像分割算法  

An Improved PCNN Image Segmentation Algorithm in Complex Background Environment

作　　者：刘军[1] 李子毅 

机构地区：[1]兰州理工大学机电工程学院,兰州730050

出　　处：《计算机与数字工程》2018年第2期375-381,406,共8页Computer & Digital Engineering

摘　　要：针对复杂背景环境下传统图像分割算法存在分割精度低、抗干扰性差等问题,论文提出一种改进型脉冲耦合神经网络(Improved Pulse Coupled Neural Network,IPCNN)图像分割算法。该算法综合考虑图像像元的灰度分布信息及像元之间的空间位置信息,在简化PCNN模型的基础之上,结合二维最大类间方差法对初始阈值进行优化,并且为了提高算法的实时性,推导并给出了相关快速递推公式;同时,不同于传统PCNN依据经验或通过大量实验确定模型关键参数的做法,而是从PCNN的耦合特性出发、结合图像自身空间和灰度特性,通过计算图像局部灰度均方差确定连接强度系数,并综合考虑像素点的空间与灰度值差异确定其连接权值矩阵,最后依据信息熵最大原则判别分割结果,实现了目标对象自适应自动分割。数字实验表明,该算法较传统PCNN算法具有图像分割速度快、目标轮廓分割清晰、抗干扰性强等优点。For the traditional image segmentation algorithm in complex background environment,there are some problems such as low segmentation precision and poor anti-interference and so on. This paper presents an Improved Pulsed Coupled Neural Network(IPCNN)image segmentation algorithm. The algorithm takes into account the gray level distribution information of image pixels and the spatial position information between pixels. Based on the simplified PCNN model,the initial threshold is optimized by combining the two-dimensional OTSU method,and in order to improve the real-time performance of the algorithm,a fast recursive formula is derived and given. At the same time,it is different from the traditional PCNN which determins the key parameters of the model by experience or a large number of experiments. The IPCNN determines the connection strength coefficient by calculating the local gray mean square error of the image based on PCNN coupling characteristics,with consider of image space and gray level characteristics. The connection weight matrix is determined by considering the differences between spatial and gray values of pixel points. At last,the segmentation result is assessed based on the maximum information entropy principle and the adaptive automatic segmentation for the target object is realized. Numerical experiments show that the proposed algorithm has advantages over the traditional PCNN algorithm in fast image segmentation,clear contour segmentation and strong anti-interference performance.

关 键 词：机器视觉 图像分割 脉冲耦合神经网络 自动分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]