A multi‐path residual network for image denoising based on edge prior information
Xuefei Bai,Yongsong Wan,Weiming Wang
DOI: https://doi.org/10.1049/ipr2.13264
IF: 2.3
2024-10-30
IET Image Processing
Abstract:To overcome the problem of the smoothness and loss of high‐frequency details at edges in image denoising, a multi‐path residual network using edge prior information is proposed. Firstly, the edge is recovered from the noisy image via the edge extraction module. After that, the edge information is input into the feature transformation network to obtain affine parameters. The affine parameters are then employed to guide the shallow feature extraction process of the main denoising network and optimize the weights of local information in the feature maps during the feature extraction process of the network. Finally, the denoised image is obtained by combining multi‐scale feature activation with residual learning. Edge‐preserving denoising is important in image analysis. However, most existing methods suffer from smoothing and loss of high frequency detail at the edges. Aiming at this issue, a multi‐path residual network using edge prior information is proposed. Specifically, the edge is first recovered by a newly designed parallel edge extraction module. Then it is used as a priori to generate affine transform parameters to guide the weight distribution between noise and edge, so that the shallow feature extraction pays more attention to preserving edges. In the deep feature extraction stage, a multi‐scale attention unit is designed to provide the spatial neighbourhood information of the feature points to filter and activate the deep features, which further enhances the ability of the network to extract fine‐grained noisy features. Finally, by taking the difference between the noisy image and the noise feature obtained with residual learning, the denoised image is produced. Extensive experiments show that the PSNR is improved by 0.00–0.78 dB on grey image denoising and 0.00–2.69 dB on colour image denoising compared with others. The denoised image has a better ability to preserve image edges, high frequency details and visual perception.
computer science, artificial intelligence,engineering, electrical & electronic,imaging science & photographic technology