柔性形态滤波的神经网络实现及遗传算法优化

柔性形态滤波是一种重要的非线性滤波方法.柔性形态滤波器的研究可分解为形态滤波运算和结构系统选择2个基本问题,一旦形态滤波运算选定后,柔性形态滤波性能就主要取决于结构系统的选择.该文提出了一种柔性形态滤波的神经网络模型及其网络参数(即滤波器结构系统)的遗传算法优化方法.进行优化训练的目的是为了使结构系统自适应地反映图像的形态结构特征,自动调整结构系统中硬核、边界和重复度的组成,从而提高对复杂噪声图像的滤波性能.实验仿真结果显示,该算法优化后的柔性形态滤波器性能得到较大改善,在收敛速度和滤波效果方面,其优化结果要优于简单的遗传算法.

Neural network implementation of soft morphological filters and optimization by genetic algorithm

Soft morphological filtering is an important nonlinear filtering method.The operation of a soft morphological filter can be divided into two basic parts that include morphological operation and structuring system(SS) selection.When morphological operations have been selected,the filter's properties depend mainly on the selection of SS.To improve results,research was done on a soft morphological filter model based on a neural network and an optimal method of its weights with a genetic algorithm.By means of optimizing training,the structuring system possesses the shape and structural characteristics of images;the compositions of hard center,soft boundary and repetition parameter are automatically adjusted.A soft morphological filter that provides good filtering result to images with complex noise can be realized.Simulation experiments show that optimized soft morphological filters are highly improved and their optimal results are better in terms of convergence speed and filtering effectiveness than images processed using a simple genetic algorithm.