RFIA 2016

I did an oral presentation at RFIA 2016 at the begining of the month.

RFIA is a french conference (held at Clermont-Ferrand this year) about pattern recognition and artificial intelligence. My presentation was on “Un modèle de décomposition pour la détection de changement dans les séries temporelles d’images RSO” and the proceedings can be found in the publication section of this website.

Paper accepted at JSTARS

My paper on Multi-temporal SAR image decomposition into strong scatterers, background, and speckle have been accepted at JSTARS.

It will feature a multitemporal decomposition model, that is able to take into account the scatterers that are present in SAR urban images. The model allows for the use of the L0 pseudo-norm for the detection of the scatterers and a prior using the Total Variation (TV) on the image. The optimum solution can be found exactly thanks to Graph-cut optimization. We also show two applications of the model: one for change detection, the other for regularization.

A draft of this paper will be soon available in the publications section.

EUSAR 2016

Last week, I went to EUSAR 2016 in Hamburg to present a method for the classification of water in SWOT images.

This method tries to take into account the variations of the class parameters due to the antenna pattern that cannot be correctly compensated due to the absence of signal in one of the classes. Therefore, we needed to develop a binary classification method, where the parameters are not constant.

You can find the slides here and the article here.

Paper accepted for an oral presentation at MultiTemp 2015

I have recently been notified that my paper about TV+L0 decomposition on multi-temporal series of SAR images is accepted to MultiTemp2015.
SAR signals are different from classical (such as optic) ones because it contains speckle and strong scatterers. This implies that we can not obtain good results with traditional classical image processing techniques on SAR images. In this paper we introduce a regularization that suits multi-temporal series of SAR images combining total variation (TV) and a pseudo-norm L0 regularization.
This model is then optimized using a graphcut technique allowing us to find the global optimum. An application of this result for change detection is presented.

You can download the paper in the Publication section!

Graph cut as an optimization technique

Back when I was studying at UPMC, I was asked to wrote a survey on optimization technique using graph cuts. This survey focus on certain problems expressed as Markov Random Fields (MRF) and show different algorithms that you can choose from depending on the number of classes in your problem, the regularity term of your model and the complexity.

Unfortunately, it is only available in French. You can find it here.

First poster!

A while ago, I presented my first poster at the Colloque “Ressources naturelles et environnement” of Institut Mines-Télécom (which would be translated as “seminar on natural resources and environment”).

This poster quickly introduces my PhD subject, and I thought you may want to check it out. Unfortunatly, it is in french. So I will write an introduction to my subject in this blog anytime soon.

Here it is: (if the viewer do not work, please click here)

Let’s go!


My name is Sylvain Lobry, and I just started a PhD in Image processing two days ago.  I wanted to write a blog about this once in a lifetime experience. I hope you’ll find it interesting!

First, I’d like to expose some of my background. I obtained an engineering degree in 2013 at EPITA specialized in science computing and images. With this degree, I obtained what I believe (or hope!) is a strong background in programmation, and I discovered some aspects about research in image processing while working at the R&D lab of my school, the LRDE, and by doing an internship in the R&D lab of Morpho. To validate this degree, I had to do an internship; I worked at a company doing financial software called Murex. While I really enjoyed the experience, I found it not fun enough compared to what I used to do at a R&D lab. So at the end of this internship, I had taken my decision: I wanted to do a PhD.

As I felt I still had a lot to learn before doing a PhD in image processing, I did a master degree in computer science specialized in image processing at the UPMC which is named M2 IMA. That’s where I met one of my supervisor named Florence Tupin who works at Télécom ParisTech, a french engineering school. Then I did an internship supervised by her and Loïc Denis on an extension of the TV+L0 decomposition to multi-temporal series of SAR images. If you’re not familiar with signal processing, that might sounds frightening! There will be a blog post where I’ll try to explain the subject and what I did during this internship.

Anyway, Florence offered me the chance to do a PhD on SAR images processing, with funding from the “Institut Mines-Télécom” and the CNES. The goal of this PhD is to find water surfaces in SAR images acquired by SWOT, a satellite that will be launched by the CNES and NASA in 2020. I will mostly work at Télécom ParisTech.

So in this blog, I’ll talk about this particular subject, but also about being a PhD student and everything I could fancy. On a personal point of view, I hope it will help me to keep a good pace during the next three years. Also, I hope you’ll find it interesting, that it will answer some questions you might have about doing a PhD and I’ll get some feedback from you.