第21卷
第3期
2013年3月
文章编号
1004-924X(2013)03-0821-07
光学精密工程
Optics and Precision Engineering
Vol. 21No, 3
Mar.2013
基于活跃度的脉冲耦合神经网络图像分割
郑欣“，彭真明
（电子科技大学光电信息学院，四川成都610054）
摘要：针对脉冲耦合神经网络(PCNN)在图像分割中需多次人工调整网络参数的间题，提出了一种基于PCNN模型的图像自动分制方法，分析了图像中影响PCNN参数设置的因素，提出了一种图像自适应分块策略。将图像划分为内部复杂程度相近的子块，克服了同一参数无法同时对图像中复杂度差异较大的不同区域准确分制的不足。利用本文提出的局部图像活跃度（ADLA)指标对不同子块自适应地确定PCNN模型参数，有效解决了传统PCNN图像分制时需要人工确定关键参数的间题。最后，采用最大二维Tallis炳准则从分割后的二值结果序列中选择最佳结果。实验表明，本文方法的分制结果轮席清晰、完整，即使在对比度低、背景呈大范圈内连续变化等复杂情况下，也具有优异的性能。与传统 PCNN分制方法相比，本文方法能自动、快速、准确地确定PCNN模型参数，且区域一致性测度（UM)，区域对比度
(CR)，形状测度(SM)，综合指标(CD等客观评价的量化指标均优于传统PCNN分制方法12%以上。关键调：脉冲耦合神经网络；自造应参数；局部图像活跃度；图像分割
中图分类号：TP391.4
文献标识码：A
doi:10.3788/OPE,20132103.0821
Imagesegmentationbasedonactivitydegree
withpulsecoupledneuralnetworks
ZHENG Xin', PENG Zhen-ming
(SchoolofOpto-electronicInformation,Universityof ElectronicScienceand
Technology of China, Chengdu 610054, China)
Correspondingauthor，E-mail;zheng_rin2@sina,com
Abstract: As Pulse Coupled Neural Network(PCNN) in image segmentation has to adjust the network parameters by manual operation for several times, this paper proposed a automatic image segmentation method based on the PCNN. The effect factors on setting PCNN parameters in an image were ana-lyzed and an adaptive segment protocol was invited to divide the images into several sub-pieces with the similar inside complexities. By which, the weakness that the same parameter could not segment exactly the region with quite different complexities in an image at the same time was overcome, Fur-thermore, the index of Activity Degree of Local Area(ADLA) proposed by the paper was used to de-termine the PCNN model parameters for different sub-pieces adaptively and to avoid the manual opera-tion for important parameter selection in the traditional PCNN image segmentation. Finally, the best result was chosen from the binary-result sequences with the max two-dimension Tallis entropy proto
收稿日期：2013-01-09；修订日期：2013-01-30.
基金项目：国家白然科学基金资助项目（No.41274127，No.40874066)，中国科学院光束控制重点实验室基金资助项
目（No.2010LBC001)；中央高校基本科研业务费专项基金资助项目（No.ZYGX2010J063）