Image processing

Research | Links:

Image and video processing is an underlying but central topic to most of my research.

I am currently leading the ENS team on the ANR Improved project, where we collaborate with the French scientific police (Service National de Police Scientifique, SNPS), EKTACOM, and the Xlim and Ceraps academic partners. The goal of this project is to create methods that restore low-quality, strongly-degraded images and videos, and derive information from them in a way that gives guarantees against hallucinations and ensures the information we derive from the images or videos is correct.

My work in forensics naturally led me to work on reverse-engineering the image processing pipeline, I thus gained expertise in the various steps of this pipeline (most importantly demosaicing, denoising and compression) and its artefacts. Most notably, I led a team effort to understand and model how JPEG compression affects prior image noise, that was already present in the image before compression. While co-advising Yanhao Li on his PhD thesis on deepfake detection, we designed a method to analyse the noise level in videos as a function of pixel luminance (see the IPOL article and demo as well).

I also coordinated a team at the ENS to analyse and create a new metric on the evaluation of HDR imagers for autonomous and assisted driving (work under publication).