Application of wavelet neural network (WNN) and gradient descent method (GDM) in natural image denoising

作者：Yan Yunyi; Guo Baolong^*

来源：Journal of Computational Information Systems, 2006, 2(2): 625-631.

• Convergence speed
• Feature normalization
• Gradient descent method (GDM)
• Image denoising
• Median filtering
• Wavelet function
• Wavelet neural network (WNN)

Denoising is still a challenging problem for researchers. Wavelet neural network (WNN) is introduced into the field of digital image denoising due to the ability of which can learn the principles to distinguish noisy points from others. In the view of approximating, the process of WNN denoising can be thought as an approximating process performed by WNN. In this paper, trained WNN recognizes noisy points with the help of nonlinear normalized feature values by an exponent function which can distinguish noise from others efficiently. The gradient descent method (GDM) is selected to train the WNN for its acceptable convergence speed. The results of experiment using Mexican hat wavelet function showed that WNN is efficient in recognizing the noisy points. The proposed approach is superior to standard median filtering in the ability of preserving fine details and excellent fidelity.

• 出版日期2006
• 单位西安电子科技大学