Real-Time Image Denoising with Embedded Deep Learning: Review, Perspectives and Application to Information System Security

Florian Lemarchand (1) , Erwan Nogues (1, 2) , Maxime Pelcat (1, 3)

1 IETR - Institut d'Électronique et des Technologies du numéRique
2 DGA.MI - DGA Maîtrise de l'information
3 IP - Institut Pascal

Any information system emits, by conduction or radiation, compromising signals likely to be intercepted by an attacker. These leakage signals usually have low signal-to-noise ratio and the security of information systems depends on the capacity of an attacker to denoise them. Denoising is a major topic in signal processing, currently revolutionized by deep learning methods. In particular, the scope of image denoising is large and ranges from classical and low footprint techniques to computationally intensive deep learning techniques. Deep learning algorithms typically run onto energy costly computers using Graphics Processing Units (GPUs) and are currently hardly available in an embedded context. This paper gives an overview of existing methods for embedded image denoising and proposes some perspectives. A case study is also presented that motivates our research on the domain.

Domaines: Intelligence artificielle [cs.AI]; Architectures Matérielles [cs.AR]; Traitement du signal et de l'image [eess.SP]; Cryptographie et sécurité [cs.CR]

Florian Lemarchand, Erwan Nogues, Maxime Pelcat. Real-Time Image Denoising with Embedded Deep Learning: Review, Perspectives and Application to Information System Security. RESSI19, May 2019, Erquy, France. ⟨hal-02082855⟩