摘要
本文从边缘点所在尺度和其邻域的灰度分布状况入手,提出了一个由四个分量组成的、对应于点的、用于衡量该点重要性的边缘重要性度量向量,为了考虑应用背景,用人工分类好的样本对一BP神经网络进行训练,用训练好的网络对图像的边缘点依重要性进行分类,从而获得图像的重要边缘。另外由于本文的方法无须对图像进行卷积,所以不会产生边缘偏移。经实验验证此方法取得了良好的效果。
Based on the scale and the state of intensity distribution in some neighborhoods of edge points, this paper proposes a measure vector of importance of point that consists of 4 components and corresponds to every point. For considering the background of application, this paper first trains a BP neural network using some samples that have classified by manual work, and then extracts more important edge points in a new image using the trained neural network. Because the image needs not be smoothed by some function in this algorithm, the edge deflection will not happen as usual, the location of gotten edge is in the accurate position. The effectiveness of this algorithm has been testified by some experiments.
关键词
边缘检测
神经网络
重要性度量
图像处理
Edge detection, Neural network, Measure of importance, Scale
