Projects
- 2009
-
Dynamic Clustering
Clustering is an important Machine Learnign task that groups data into a limited number of classes with respect to their similiarities and dissimilarities. Behind these notions, one has to choose an appropriate distance for the data so the aim of this project was to define dynamically and in a adaptive way a distance that fits the data to ease classic clustering algorithms such as K-Means, Spectral Clustering. I would to thank Hani HAMDAN with whom I had the real honor to work with. |
Matlab for the Monte Carlo's simulations |
-
Wavelet Image Denoising
Denoising image with wavelets is a very well-known technique. By the way, this project tries to put a link between wavelets' thresholding and hysteresis as in Canny's edge detector. As wavelets have non zero support, when we have a singularity, a wavelet needs some space to
express itself. If we cut the tails of the wavelets, we get a Gibbs effect or the ringing effect appeared because of the negative of the tails that lacks. So the strategy is to select seeds and let them grow like in Canny's detector. |
Matlab |
- 2008
-
Detection and description of key-interest points
To compare two images, one strategy is to detect interest points (corners, blobs ...) and analyse their neighborhood. The goal of this project is to build a new descriptor of these interest points by a Machine Learning approach. This is implemented by using SURF with the help of Renaud KERIVEN, Jean-Yves AUDIBERT, Francis BACH and of course my class mate Olivier DUCHENNE. |
C++ (OpenCV on Visual Studio), Matlab |
- 2007
-
Denoising images
Denoising images is a very old scientific challenge. This is a sine qua non step for further applications in Computer Vision. This work was about the fast implementation of the Fast Bilateral Filter of Frédo DURAND and Sylvain PARIS. |
Matlab, C (MEX) |
-
Stereoscopy
Take two photos and then compute the tridimensional coordinates of it. This work was done under the supervision of Philippe MOROSINI and Jean-Louis GUTZWILLER. In fact, my knowledge about this topic was very poor and constrained by limited time in spite of Hafed TOURMIHINE and Julien KELDER's hard work. |
Matlab |
- 2006
-
Monopoly
Numerical determination of the best Monopoly's artificial player in a very simple model with a Genetic Algorithm and a Distributed Architecture equipped with a Dynamic Load Balancing system. The numerical results of Java simulations are then read graphically on Matlab. Hafed TOURMIHINE behaved like the real engine of our work supervised by Stéphane VIALLE. |
Java, Matlab |
-
Speaker Recognition
Speaker identification and verification with Matlab and DTW (Dynamic Time Warping from Dynamic Programming), Cepstral Analysis and Probability. The results are then dramatically improved by using a probability model which is a smoother criterion of discrimination than a simple DTW distance threshold between the speakers. In this project, my friends were Grégory CARRÉ and Jonathan JACQUIN. |
Matlab |
- 2005
-
Distributed Garbage Collector
Development of a Garbage Collector inside the NGrid project, a distributed computing environment for genetics and its recent improvements, within a 9-scientist team at the Center for Computational Biology of les Mines de Paris. This project is fundamentally based on Graph Theory. This work was born in Joannès VERMOREL's hands and also developped by Grégory CARRÉ and Jonathan JACQUIN. |
C#, Java (for the tests) |
-
Scoring Machine
Off-line reconstruction in Java of music scores from a .wav sound file while trying to define a new but efficient statistical way to determine the tempo. This project mainly uses Spectral Analysis and Butterworth Filters without knowing anything about Wavelets and Multi-Scale approaches. Philippe ESTÈVE was my buddy. |
Java, Matlab (for the tests) |
|
