小波神经网络在CCD图像噪声处理中的应用

目的：CCD相机响应功能的非线性，导致了CCD噪声模型的复杂性，使得滤波效果不佳，本文提出一种针对于数字图像中CCD噪声的小波神经网络滤波器。方法：首先，分析CCD噪声模型，找出导致CCD噪声模型复杂的原因——CCD相机响应功能（camera response function简称CRF）的非线性；接着，在对ANS滤波器分析的基础上，针对影响滤波效果的两大问题：滤波窗口和图像强度，将小波神经网络非线性逼近CCD噪声曲线，按照噪声参数对图像进行区域划分并分配相应的权值，然后结合相应的非线性滤波器进行针对性滤波，最后综合输出。结果：实验结果表明：本文改进的滤波器滤波效果明显，信噪比得到进一步提高（24.65）。结论：利用神经网络良好的非线性函数逼近性，将其结合ANS滤波器构造出神经网络非线性ANS滤波器（NN-NANS filter），试验结果表明，该滤波器在去除噪声的同时边缘细节也得到了很好的保留，同时提高了信噪比。

Application of wavelet neural network in removing CCD noise in digital images

Objetive: Nonlinearity of camera response function makes for complexity of CCD noise model and makes the traditional filter inefficient, so a wavelet neural network filter is proposed to remove CCD noise in digital images. Method: First, by analyzing the CCD noise model we find out the reason for the CCD noise model’s complexity---nonlinearity of camera response function and image intensity. Second, on the basis of analyzing ANS filter, two factors affecting the filter---filter window and image intensity are found. wavelet neural network (WNN) is used to approach the photon transfer curve (PTC) , classify the image according to the noise parameter and assign the coefficient, and then suited nonlinear filters are used to remove noise , at last all filters’ output are combined for the final output. Result: Experimental results indicate that WNN-NANS filter has a better filtering effect and increases the SNR（24.65）. Conclusion: Because of good approach to nonlinearity, neural network is combined with nonlinear ANS (Adaptive noise smoothing) filter for a new filter-WNN-NANS (neural network-nonlinear adaptive noise smoothing) filter. Experiment results show that the WNN filter is more efficient in noise removal, edge reservation and SNR.