Abstract: machines which are one of the best
Abstract: Given the ever growing amount of visual content available on the
Internet, the need for systems able to search through this content has grown.
based image retrieval systems have been developed to address this need. But
the growing size of the databases, new challenges arise. In this thesis, the ne
grained classification problem is studied in particular. It is first shown that
techniques, and in particular the support vector machines which are one of the
image classification technique, have some difficulties in solving this problem.
often lack of exploration in their process. Then, evolutionary algorithms are
considered to solve the problem, for their balance between exploration and
exploitation. But their performances are not good enough either. Finally, a
hybrid system combining an evolutionary algorithm and a support vector machine
is proposed. This system uses the evolutionary algorithm to iteratively feed
the support vector machine with training samples. The experiments conducted on
Caltech-256, a state of the art database containing around 30;000
images, show very encouraging results.
Keywords: CBIR, Evolutionary Algorithm, SVM, Fine Grained Classification
First of all, thank you to both of my supervisors Denis Pallez and
without whom this document would have never exist. They supported me throughout
the whole process to complete this thesis, helped me overcome the difficulties
and were very supportive with regards to the personal problems I encountered
during the thesis. They believed I could achieve great things all along and I
I have done is enough to fulfill at least part of their expectations.
Thank you Christian Brel, for all the help and support he gave me
when I most
needed it. I couldn’t have went through the personal issues I faced so easily
his help, and he has always been supportive.
Thank you Magali Brel, for all the love and support you gave me
two past years. Without you and Christian, I probably would have given up at
Thank you to my family, they were always there to support me, in
my mother, Corinne, and my sister, Anaïs, who helped me recover when I needed
it. I always enjoy the time I spend with all my family and the vacation I had
chance to take during the thesis to visit them helped me overcoming the difficulties
Thank you Paul Sachez, a long time friend who always supported me
me overcome all the difficulties I encountered. Thank you to his parents too,
kindly accepted to host me for a few days when I really needed it.
Thank you Dimitri Martinho, another very long time friend who has
supportive and came to spend some time with me. He has always been there when
I needed and I hope I’m there for him too.
Thank you Anne-Marie, Jean-Paul, Ivan, Cyril, Christian Delettre,
Mélanie, Katty, Yoann, Sébastien, Simon, Sami, Clément, Cécile and all the
people from the SPARKS team I had the chance to talk with and spent time with.
Speaking with all of you was a great pleasure, I truly enjoyed all the time I
with each of you. The moments we spent together, either at lunch, somewhere in
the afternoon, an evening around boardgames or dining at a restaurant, or even
working together were great and helped me in completing this thesis.
Thank you Magali Richir, Régine Saelens and all the administrative
the I3S laboratory and the doctoral school STIC from Nice for their help and
support. Doing all the administrative tasks from afar while being in recovery
have been possible without your help. And thank you for the help and support
provided me all along the way.
Thank you to all the friends from our teamspeak server: Aerendil,
Krusk, Shysu, Mimil, Kira, Ursakar, Mathosse to name just a few. I spent a lot
time playing and speaking with you all, which helped me a lot too. Also thank
Riot Games, Quickshot, Krepo and all the league of legends casters and
playing the game and watching the pro scene was part of the distraction that
me escape from the work for some time to better get back to it after.
Thank you Cyril Fonlupt and Arnaud Revel for accepting to review
and thank you Carlos Artemio Coello Coello and Andrea G. B. Tettamanzi for
accepting to be member of the jury of my PhD defense. I hope you will
I’d also like to thanks Florent Perronnin for giving us the image
used in his work. Those would have taken several weeks to obtain without his
Finally thank you, the reader, for reading this. It means a lot to
me, because it
means that you find this work is worth investing some of your time, and
the time we spent with my supervisors to create it, I hope you will find
interesting in it