摘要
针对文献[1]中提出的CHNN图像边缘检测算法缺乏足够的参数来调节边缘检测的灵敏度以及检测结果图像边缘过宽的缺陷,提出一种改进的CHNN方法,称之为Weighted CHNN(加权的CHNN,简称WCHNN)方法。该方法在CHNN神经网络元的n个连接上施加权值,可以通过各种局部搜索、优化算法,使用指定的样本输入、样本输出等方法来训练该WCHNN网络从而确定各权值,使得WCHNN在保留了CHNN的优点的同时,还可以根据不同的样本输入输出图像来调节边缘检测的灵敏度,从而提高检测结果质量并避免检测结果中出现边缘过宽的情况。实验结果表明,训练后的WCHNN网络,比起CHNN有着更低的边缘检测错误率,并可检出原来CHNN方法漏检的边缘。
In response to the two flaws of CHNN image edge detection algorithm brought up in literature [1] that on the one hand there are not enough parameters to adjust the sensitivity of edge detection,on the other hand the image edges as detection results are too wide,an improved CHNN method called Weighted CHNN（WCHNN for short） is proposed.The method imposes weight values on n connections of CHNN neural network neurons to apply such methods as local search,optimized arithmetic,assigned sample input and output etc.to train WCHNN network in order to determine each weight value,so that WCHNN preserves the merits of CHNN while depends on different sample input/output images to adjust the edge detection sensitivity so that not only the quality of detection result is improved but also the occurence of over wide edge in detection result can be avoided.Experiment results prove that,compared with CHNN,the trained WCHNN network achieves lower edge detection error rate;moreover it can find out edges which may be neglected by the former CHNN method.
出处
《计算机应用与软件》
CSCD
2011年第5期255-258,267,共5页
Computer Applications and Software
关键词
图像边缘检测
CHNN
人工神经网络
加权参数
参数训练
Image edge detection CHNN Artificial neural network Weighted parameter Parameter training
