Designing and training of a dual CNN for image denoising

Deep convolutional neural networks (CNNs) for image denoising have recently attracted increasing research interest. However, plain cannot recover fine details a complex task, such as real noisy images. In this paper, we propose Dual Network (DudeNet) to clean image. Specifically, DudeNet consists of four modules: feature extraction block, an enhancement compression and reconstruction block. The block with sparse mechanism extracts global local features via two sub-networks. gathers fuses the provide complementary information latter network. refines extracted compresses Finally, is utilized reconstruct denoised has following advantages: (1) dual can extract enhance generalized ability denoiser. (2) Fusing salient (3) A small-size filter used reduce complexity Extensive experiments demonstrate superiority over existing current state-of-the-art methods. code accessible at https://github.com/hellloxiaotian/DudeNet.