一种基于神经网络的形态滤波器优化设计方法

Design of optimal in morphological filters based on neural nets

给出了一种利用神经网络实现灰度形态滤波器参数的优化设计方法。文中首先简述了形态滤波器的基本理论,并设计出了灰度形态学的两种最基本运算(膨胀和腐蚀)的神经网络结构,其网络权值即为形态运算的结构元素值。然后,按照d学习规则,自适应地对结构元素(形状和数值)进行学习训练,以获得最佳的灰度形态滤波器参数。该方法设计简便,实用性强且易于推广,对提高形态滤波性能效果显著。

The paper presents a novel algorithm to design optimal parameters of a gray morphological filter with neural network. Two neural network structures of dilation operation and erosion operation are giren. Among them, weights of neural network are represented by a structuring element and trained by a learning algorithm based on d-learning criterion. The results of numerical examples will be shown that the algorithm has better filter properties than the conventional morphological ones.