|Since 2016||Associate Professor||Université de Haute-Alsace|
|2013-2016||Head of Multimedia and Web Department||Saint-Dié Institute of Technology|
|2012-2016||Associate Professor||Université de Lorraine|
|2011-2012||Teaching and Research Assistant||Université de Lorraine|
|2010-2011||Teaching and Research Assistant||Université de Strasbourg|
|2007-2010||R&D Engineer||Ready Business System|
|2011||PhD in Computer Science||Université de Strasbourg|
|2007||MSc in Computer Science||Université de Strasbourg|
My research work on template-matching is mainly focused symbol spotting in technical documents. In such document, the spotting is hard because it suffered from the overlapping of symbols. So, the template-matching operator has to be robust to information overlap which eliminate most of the classical operators. To overcome this issue, we design a new hit-or-miss transform operator able to deal with information overlap. I also develop softwares to illustrate the efficiency of our proposed approach. Currently, I am working on the fully automatization of the method through machine learning approaches and its application to new data types such as bill tickets. I have also worked on template-matching applied to remote sensing images.
In the context of the CHIST-ERA AMIS project I will work on overlaid text localization in news video.
Segmentation is often the first step of image understanding process. Our contributions in this domain are various but all based on interactive quasi-flat zones segmentation method. Initially designed for video segmentation, we work on the application of this method to different image types and problems such as image filtering, satellite images time-series classification and tree leaves segmentation.
I also participate in the supervision of the PhD thesis of Rachida Es-Salhi on the visual data database co-segmentation which is typically a research subject dealing with all the problems induced by the data deluge.
As mathematical morphology needs ordering, its application to color image is not straightforward. In fact, there is no generic optimal vectorial ordering able to deal with every image processing problems in all color spaces. In this context, we study existing vectorial ordering, design new ones and compare them through different image types and applications. We also work on the adaptation of existing morphologic tools to color image.
|APIM||PIR UHA||2017||Project Leader|
|Bastien Latard||2016||ongoing||PhD||Semantic Analys of scientific articles|
|Oleg Eremin||2017||ongoing||Master||Deep Learning applied to medical imagery|
|Rahul Sahal||2017||ongoing||Master||Overlaid text extraction in news video|
|Giovanni de Angelis||2016||3 months||Master||People counting using 3D camera|
|Antonio Terrone||2016||3 months||Master||People counting using depth image|
|Roberto Pisapia||2016||3 months||Master||People counting using embedded systems|
|Julien Bidolet||2014||6 months||Master||Graph indexing for image classification|
|Michał Kowalczyk||2012||5 months||Master||Image segmentation on mobile environment|
|Jean-François Kraemer||2010||4 months||Master||Video segmentation/annotation tool|
|Vincent Danner||2009||2 months||Master||Optimized video management for Pelican|
Polyvalent Extensible Library for Image Computing and ANalysis is a multi-platform framework, written in Java, dedicated to Image Processing. The project started at the Université de Strasbourg in 2005 but its contributors are no longer in this university. It allows the processing of image from different types (2D, video, 3D, 3D+t) and origins (casual, medical, astronomic, remote sensing, ...) and contains many standard image processing algorithms.
Pelican contains more than 100k lines of code and is freely available under CC BY-NC 4.0 license.
Online application of a previous version of TeSySp using JavaServer Pages technology. It allows to easily and quickly test the symbol spotting approach presented at ICPR 2012.
Software developped during the ANR-JC ECOSGIL project. Written in Java, it is a tool dedicated to geographers. It achieves several processing on coastal remote sensing images such as the extraction of coastline. It uses the Pelican framework. It is still used by geographers (initial release dated from 2007).
TeSySp (Technical Symbol Spotter) was initially developed to demonstrate our work on symbol spotting in technical documents. Written in Java, it is based on Pelican framework. It is an improvment of our initial approach, in particular the use of machine learning techniques to filter false-positives result. Moreover, TeSySp allows to create ground-truth for symbol spotting and contains different metrics to evaluate the quality of spotting results.
|Since 2016||Computer Science Department||École Nationale Supérieure d'Ingénieurs Sud-Alsace|
|2011-2016||Multimedia and Web Department||Saint-Dié Institute of Technology|
|2013-2016||Computer Science Department||Ecole Nationale Supérieure des Mines de Nancy|
|2009-2011||Computer Science Department||Robert Schuman Institute of Technology|
|2009||Physical Measurements Department||Louis Paster Institute of Technology|