Deep Residual Learning for Burst Denoising
Hanlin Tan, Huaxin Xiao, Shiming Lai, Yu Liu, Maojun Zhang · 2019
The goal of image denoising is to recover a clean image from noisy input(s). For single image denoising, we look for similarities (or priors) within and across an image dataset to help recover clean images. As noise level increases, using multiple frames become feasible, which is defined as burst denoising. In this paper, we propose a deep residual model for burst denoising. Unlike previous methods, our model does not need an explicit aligning procedure, which is light-weighted and fast. Since denoising performance is closely related to noise level, frame displacement and number of frames (burst length), intensive experiments and ablation study are performed. Results show that the proposed method performs significantly better than previous state-of-the-art methods V-BM4D and KPN in all valid test cases.