MODEL - Learning Joint Denoising & Demosaicing for Mobile Devices

The main aim of this project is to enhance the quality of pictures captured with smartphone cameras by developing new imaging methods for joint denoising and demosaicing. We use a learning approach based on variational networks that both retains the interpretability of variational methods and at the same time profits from the efficiency and excellent performance of deep learning methods. Key aspects we will focus on include the preparation of a groundtruth database, the incorporation of noise statistics and video processing.