Combining multi-level loss for image denoising

The image processing has witnessed remarkable progress in image denoising. Nevertheless, restoring the visual quality of the image remains a great challenge. Existing methods might fail to obtain the denoised images with high visual quality since they ignore the potential connection with the high-level feature and result in over-smoothing results. Aiming to research whether high-level feature could influence the performance of denoising task, we propose an end-to-end multi-module neural network architecture, which introduces the combination of the high-level and low-level feature in the training process, for image denoising. In order to guide model preserve structural information efficiently, we introduce a hybrid loss, which is designed to restore details from both pixel and feature space. The experimental results show our method improves the visual quality of images and performs well compared with state-of-the-art methods on three benchmarks.