Jean-Marc Berthommé

PhD student in Computer Vision,

Université Blaise Pascal,
Institut Pascal,
ComSee - ISPR.

The aim of my PhD is to investigate non-parametric models for object tracking and recognition through
three key questions :

- How to update a model along the time ?

- How to model a probability density function in a high dimensional space ?

- How to model a probability density function from a very high number of points ?

My subject is available here (french version) and here (english version).

- In Proceedings of VISAPP 2013. Preprint & Poster.
- Simulate a well-known optical illusion due to the blind spot.
- Fill-in any binary, grayscale or color images thanks to the entropy inpainting principle.

- In Proceedings of VISAPP 2012. Preprint & Poster.
- Learn a set of kernels that fit to any labelled ROI by CMIM.
- Track a ROI with a MCMC Particle Filter and compare different sampling strategies.

- Another CMIM code to retrieve the features that fit to any image labelling.
- Note that a good feature can either belong to the object or not as H(Y) = H(~Y).

- CMIM algorithm to select few half-planes that separate binary points.
- Read F. Fleuret's paper and check D. MacKay's famous book (Chap. 8).

- CCVT algorithm to balance the capacities of 3 sites.
- Read Capacity-Constrained Voronoï Diagrams in Finite Spaces.

- Lloyd's algorithm on a uniform pdf to cluster points.
- Look at Capacity-Constrained Voronoï Tesselation (CCVT) too.

- C Programming/Beginning exercises based on Alexandre Guitton's course (2009/2010).

- Control (french) based on Thierry Chateau and Lounis Adouane' course (2012/2013).

- Computer Vision projects (2D & 3D) with engineering physics students (2012/2013).

Supervisor : Michel Dhome

Co-supervisor : Thierry Chateau

Institut Pascal - UMR 6602

24, Avenue des Landais

63177 AUBIÈRE Cedex

FRANCE