Blind image denoising method based on wavelet autoencoder

In order to solve the problem that the depth learning based image denoising method has limited denoising ability and poor denoising effect in the blind image denoising task,a wavelet autoencoder (Wavelet-AE) was constructed and it was applied to blind image denoising．Firstly,Haar wavelet was used to construct wavelet convolution layer (DWT) and inverse wavelet convolution layer (IDWT)．Secondly,DWT and IDWT were used to build down sampling block and up sampling block respectively．Finally,a wavelet autoencoder was constructed using down-sampling blocks and up-sampling blocks．Wavelet-AE can establish the mapping relationship between noisy images and clean images in the feature space,thus it has better ability for blind image denoising．To quantitatively evaluate the blind image denoising ability of the proposed method,the peak signal to noise ratio and structural similarity were used as evaluation indicators．The experimental results of image blind denoising on Kodak and SIDD datasets show that Wavelet-AE has better blind denoising ability than some advanced traditional methods and depth network denoising methods．The average peak signal to noise ratio (PSNR) can be increased by 4.58 dB at most,and the average structure similarity index measure (SSIM) can be increased by 0.149 at most． © 2024 Huazhong University of Science and Technology. All rights reserved.