Image denoising based on deep feature fusion and U-Net network

Image noise hinders the understanding of images by advanced visual tasks, and removing image noise is a challenging task. The traditional denoising methods can not only destroy the texture of the image, but can not save the image texture after removing the noise. Therefore, we propose a novel image denoising method based on deep feature fusion and U-Net network. This new method uses a two-branch U-Net network to fuse features and preserve image texture. In this paper, two encoders with independent parameters are proposed to extract more useful information respectively, and a fusion module with series connection is proposed to make better use of the extracted information and remove redundant information. Finally, the decoder is used to reconstruct the image, and the U-Net peer connection is used on the symmetric convolutional layer in the network. A large number of experimental results show that the proposed algorithm can effectively remove synthetic noise and real noise, and the reconstructed image has a good effect on both subjective visual effect and objective evaluation index.