融合多算子信息的BPNN边缘检测研究  

Research on BPNN with Fused Information from Multi-Operators for Edge Detection

作　　者：芶盛[1,2] 王茂芝[1,2] 

机构地区：[1]成都理工大学信息管理学院,成都610059 [2]数学地质四川省高校重点实验室,成都610059

出　　处：《微计算机信息》2010年第17期209-211,共3页Control & Automation

摘　　要：边缘检测是图像分割、模式识别等的基础,其实质是一个分类问题。本文综合利用微分和中值滤波算子信息,构建了像素点的9维特征向量,同时综合了Canny算子和LOG算子的优点用于选取网络学习样本,进而构造BP神经网络用于图像边缘检测。实验结果表明:本文构建的BP神经网络能够检测连续性弱边缘,可以有效提取图像边缘信息,同时网络具有较强的鲁棒性。最后指出了进一步的研究内容。Edge detection, whose essence is a classification problem, is the foundation for image segmentation, pattern recognition, etc. This paper utilize the information of differential and filter operator comprehensively, construct nine-dimensional feature vector for the pixels of the image ,and integrate advantages of Canny and LOG operator during selecting the network samples, then construct the BPNN for image edge detection. The experiment indicates that the BPNN proposed in this paper can detect continuous weak edge and extract the image edge effectively and the network has very strong robustness. At last, the future research work has been discussed.

关 键 词：边缘检测算子 BP神经网络 阈值 鲁棒性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]