Erik Bergrahm and Max Sjölin presents their master’s thesis Leveraging IIR State in Deep Learning Neural Networks for Video Denoising Wednesday 28 May at 15:15 in MH:227

Abstract:
Machine Learning models have in recent years changed the field of video denoising. Most state-of-the-art models utilize multiple frames as input to denoise a single frame, which limits the temporal context window and often requires more computational power. Another, less explored, approach is to use recurrence to gain a hidden memory state in which an additional temporal state is stored. While the former approach scales poorly with larger temporal context windows, the latter lacks proper analysis of its relevance and trade-offs. In this master’s thesis, we establish experiments to better understand the feasibility of using recurrence for these tasks, and to study the artifacts that this approach can introduce. Furthermore, we attempt to construct objective metrics to measure the flickering, ghosting, and trailing, that temporal filters can give rise to. We use a dataset of statically mounted surveillance cameras, upon which we add synthetic noise using a realistic noise model. With the constructed metrics, as well as the standard performance measures, we present differences in using these models. We also compare these models against a state-of-the-art traditional video denoising model, together with a machine learning model performing solely spatial noise filtering. The results show that the recurrent model outperforms the other models in denoising the video, while preserving the details and temporal consistency in static regions. This comes at the expense of worse performance in objects that move across the scene. The presented metrics also show promise in objectively capturing effects that standard metrics miss.

Examiner: 
Alexandros Sopasakis, Centre for Mathematical Sciences, Lund University

Supervisors:
Kalle Åström, Centre for Mathematical Sciences, Lund University
Gustav Hanning, Centre for Mathematical Sciences, Lund University
Andreas Muhrbeck, Axis Communications
Fritjof Jonsson, Axis Communications
Nils-Erik Olofsson, Axis Communications