摘要
本文构造了不同的信息测度来定量描述阶跃边缘的三个本质特征 ,并给出由相应的三个分量组成的特征向量 .用人工得到的样本对一BP神经网络进行训练 ,将训练后的神经网络直接用于图像的边缘检测 .本文方法无需定阈值 ;在特征的选取上充分考虑了边缘和噪声的本质区别 ,具有优异的抗噪性能 .实验证明本文方法具有令人满意的效果 .
In this paper,from the natural characters of step edge of image,we proposed a feature vector that consists of 3 components based on the information measure accordingly.Firstly.a BP neural network is trained with some samples that have been classified by manual work,and then extract edge proints in a new image using the trained neural network.Our method does not need any threshold and has better anti noise performance since the influence of noise is adequately considered when the feature vector is settled.The effectiveness of this algorithm has been testified by some experiments in this paper.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2001年第1期51-53,共3页
Acta Electronica Sinica
