Selective Residual M-Net for Real Image Denoising

However, these complex architectures cause the Image restoration is a low-level vision task which is to restore restoration models to waste more computation and the improvement degraded images to noise-free images. With the success is only a little. of deep neural networks, the convolutional neural networks In this paper, we try to balance between the accuracy surpass the traditional restoration methods and become the and computational efficiency of the model. First, we propose mainstream in the computer vision area. To advance the performance the hierarchical selective residual architecture which is based of denoising algorithms, we propose a blind real image on the residual dense block with a more efficiency structure denoising network (SRMNet) by employing a hierarchical named selective residual block (SRB). Moreover, we use the architecture improved from U-Net. Specifically, we use a multi-scale feature fusion with two different sampling methods selective kernel with residual block on the hierarchical structure (pixel shuffle [18], bilinear) based on the proposed M-Net called M-Net to enrich the multi-scale semantic information.