A lightweight convolutional neural network for image denoising with fine details preservation capability

Image denoising is a fundamental problem in image processing whose primary objective is to remove the noise while preserving the original image structure. In this work, we proposed a new architecture for image denoising. We have used several dense blocks to design our network. Additionally, we have forwarded feature extracted in the first layer to the input of every transition layer. Our experimental result suggests that the use of low-level feature helps in reconstructing better texture. Furthermore, we had trained our network with a combination of MSE and a differentiable multi-scale structural similarity index(MS-SSIM). With proper training, our proposed model with a much lower parameter can outperform other models which were with trained much higher parameters. We evaluated our algorithm on two grayscale benchmark dataset BSD68 and SET12. Our model had achieved similar PSNR with the current state of the art methods and most of the time better SSIM than other algorithms.