基于RBF神经网络的交叉检验图像降噪方法

Image Denoising Using Cross-Validation Method with RBF Network Representation

径向基神经网络(RBFNN)有很好的函数逼近能力,本文把图像看成函数,用RBFNN来表示图像,并在学习过程中进行逼近,采用交叉检验的方法来确定网络模型并控制学习过程,使得网络既能较好地逼近图像,又对噪声进行抑制,从而达到图像降噪的目的。实验证明这是一个可行的盲目图像降噪方法,有广泛的适用性。本文还给出了实验中改进交叉检验方法的技术,给出了与wiener滤波器降噪的实验效果对比图。

This paper presents an image denoising method based on RBF neural network (RBFNN), shows the method's advantages and disadvantages and gives a comparison between this method and other classical image denoising methods. RBFNN is good dat function approximation, especially approximating to smooth functions. Experiments show that RBFNN can well approach the image while repressing the noise.