一种新的强鲁棒性的边缘检测方法  

A New Robust Method of Edge Detection

作　　者：张晶[1] 张权[1] 王欣[1] 

机构地区：[1]山东大学信息科学与工程学院,济南250100

出　　处：《信号处理》2007年第3期455-459,共5页Journal of Signal Processing

基　　金：国家自然科学基本项目(60172022);山东省自然科学基金(Z2005G02)

摘　　要：针对图像同时叠加脉冲噪声和高斯白噪声的非标准分布噪声的情况,本文提出一种新的基于统计向量和神经网络的强鲁棒性的边缘检测方法。首先选取窗口子区域内若干中间值像素点构造了由4个统计量组成的统计向量。然后计算训练图像的统计向量作为样本,对不加噪的训练图像的统计向量降维并作双阈值处理得到学习边缘图,对BP神经网络训练。最后将训练的BP神经网络直接用于边缘检测。新方法对脉冲噪声和高斯白噪声均具有较好的鲁棒性,BP神经网络的结构和训练都比较简单,而且不需要设定阈值检测边缘。A new robust edge detector based on the statistical vector and neural network is proposed in the paper,which aims to decrease a mixture of impulsive and Gaussian noise whose distribution departs from normality. Firstly, a statistical vector composed of 4 components is presented,for which several medial pixels in a window are selected as samples. Then through the training with statistical vector samples calculated from training images and a teaching edge map,the BP neural network acquires the function of a desired edge detector. The teaching map is got by applying de-dimension and hysteresis thresholding to statistical vectors of the noiseless training image. At last,the trained BP neural network is used for edge detection directly. The proposed edge detector proved robust against both impulsive and Gaussian noise in experiments. Besides, both the architecture and training of the BP neural network are simple. Moreover, the proposed edge detector needs no thresholds for conventional edge detection methods.

关 键 词：边缘检测 统计向量 BP神经网络 鲁棒性 双阈值 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]