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 shortcomings of CHNN image edge detection algorithm brought up in literature'1' that it lacks enough parameters to adjust the sensitivity of edge detection and that the edge detected is too wide, an improved CHNN method which is called Weighted CHNN (WCHNN in short) is presented. This method puts weights on n connections of each element of neural network of CHNN, and trains the WCHNN network using various local search and optimisation algorithms, as well as using designated sample input and output images to determine every weight value, therefore the WCHNN gains the ability of tuning the sensitivity of edge detection according to different inputs and outputs of images while retaining the advantages of CHNN, so that improves the quality of detection results and prevents the situation of too-wide edges in detection. Experimental results show that the trained WCHNN network can get lower edge detection error rate than CHNN, and it can also detect those edges missed by CHNN.
展开▼
