Image processing

Stripes / fringes denoising

This work was motivated by a collaboration with IP3D team from ITAV. The objective was to denoise images issued from a light sheet microscope (SPIM) build in Toulouse. The model that was proposed and the algorithm that was implemented can also apply to images issued from other modalities, as it can remove stripes (or fringes).

Our contribution: We have proposed a noise model adapted to the stripes that impair SPIM images. It is a stationnary noise, but it is not pixelwise independent (a stripe affects several pixels !). We have provided a variational problem expressing the denoising problem, using a Maximum A Posteriori principle. We have also developed an algorithm that is distributed as a FIJI plugin.

Results obtained on SPIM images:


Left: original image (zebrafish embryo); right: restored image


Left: original image (multicellular tumor spheroid); right: restored image

other results

The same algorithm was applied to AFM images (data courtesy JM Pellequer)


Left: original AFM image; right: restored image

More details:

Variational Algorithms to Remove Stationary Noise: Applications to Microscopy Imaging

A new scheme for anisotropic diffusion in image processing

A new scheme is developed in collaboration with Jean-Marie Mirebeau (CEREMADE, Paris Dauphine). This scheme proposes a discretization of anisotropic diffusion operators that is based on an adequate choice of neighbors based on Lattice Basis Reduction (instead of the standard 9 points stencil).

Results obtained for coherence-enhancing diffusion:

Coherence-enhancing diffusion is an anisotropic diffusion process where the diffusion tensor depends on the image structure, and the diffusion tensor is locally along the main direction of the gradient of the image.


From left to right: original image; image obtained using our scheme; image obtained using a finite difference scheme; image obtained using a finite elements scheme.