神经网络噪声检测的自适应加权均值滤波算法

摘要

针对传统滤波方法对纹理比较细腻的图像以及高噪声密度图像的处理能力欠佳的缺陷,提出了一种基于BP神经网络噪声检测的自适应加权均值滤波方法.用训练好的BP神经网络检测出图像中被椒盐噪声污染的像素并对其进行标记,对检测出的噪声点进行自适应加权均值滤波,信号点则保持不变,从而实现了对图像细节的有效保护.仿真表明了该算法滤波性能和细节保护能力均优于各种传统滤波算法.%Aimed to the flaw that the traditional filtering methods have a poor handling capacity to fine texture and high density noise,a new adaptive weighted mean filtering method based on BP neural network is proposed to remove the salt & pepper noise and protect the details of images.The salt & pepper noise of images is located via the well-trained BP neural network firstly.Then they are removed by adaptive weighted mean filtering algorithm while nothing to do with the uncontaminated pixels.Thus the details of the image are well protected.Compared with other denoising methods,the experimental results show that the proposed algorithm has a decisive advantage over all kinds of traditional filtering algorithms in filtering performance and detail protection.