Joint Denoising and Demosaicking With Green Channel Prior for Real-World Burst Images

Denoising and demosaicking are essential yet correlated steps to reconstruct a full color image from the raw filter array (CFA) data. By learning deep convolutional neural network (CNN), significant progress has been achieved perform denoising jointly. However, most existing CNN-based joint (JDD) methods work on single while assuming additive white Gaussian noise, which limits their performance real-world applications. In this work, we study JDD problem for burst images, namely JDD-B. Considering fact that green channel twice sampling rate better quality than red blue channels in CFA data, propose use prior (GCP) build GCP-Net JDD-B task. GCP-Net, GCP features extracted utilized guide feature extraction upsampling of whole image. To compensate shift between frames, offset is also estimated reduce impact noise. Our can preserve more structures details other removing Experiments synthetic noisy images demonstrate effectiveness quantitatively qualitatively.