Research

Research Interest

  • Signal & Image Processing.

  • Applied Mathematics.

  • Machine Learning & Pattern Recognition.

  • Statistical signal processing.

Ph.D. Thesis

New statistical modeling of multi-sensor images with application to change detection

  • Abstract: This Ph.D. thesis aims at evaluating the interest of using multivariate distributions for the analysis of heterogeneous images. The considered heterogeneous data are composed of images acquired by different sensors (including optical, radar and hyperspectral sensors) and possibly of an object database (containing roads, building, etc.). The applications considered in this thesis are mainly image registration, change detection and database updating. All these applications require to define a similarity measure between the different images or between features estimated from these images (called modalities).
  • Advisors: Frédéric Pascal, Jean-Yves Tourneret, Marie Chabert, Alain Giros
  • Partners: This thesis is conducted in partnership with the CNES.
  • Download: pdf (not final version), presentation

Previous research

  • 2008 - 2012

    Speaker Verification

    Areas: Biometric Recognition, Voice Processing, Signal Processing

  • 2011

    Depth from focus

    Project Tutoring

    Areas: Image Processing,
    Learn more
    This project consisted on estimating the depth of different objects from a pair of image taken from the same angle, a shallow depth of field and different focal plane.
    The main areas of this project involved modeling the Point Spread Function (PSF) introduced by the lens, estimating the model parameters on the different regions of the image using the Discrete Fourier Transform, and estimating the object depth from these parameters. Limitations arising from the non injective nature of the PSF as a function of the depth were also studied, leading to the proposal of an optimal camera configuration based on the desired depth range.
  • 2009 - 2010

    Three-phase Switching Power Factor Corrector

    Areas: Power Electronics,
    Learn more
    The objective of this project was designing a power factor correction device using switching technology, as well as analyzing its time between failures, cost and commercial viability.
    The system consisted of a measurement unit, a control unit and a switching current management unit. The measurement unit measures the instant voltage and current consumption using hall effect sensors. This information was then used by the control unit to compute the instantaneous current difference between the actual consumption and the consumption of a pure resistive load based on the dqo transform. Then the current management unit injected the current difference into the system through a switching current power supply to compensate the actual load consumption. The system designed was able to compensate the power factor arising from either inductive, capacitive or non linear loads.
    The results of this project were presented on the ITBA fair of electronics, where it was chosen as the best R+D project by four of the fair sponsors.
  • 2009

    8051 Dualcore Pipeline Processor

    Areas: Microprocessors Architecture, FPGA,
    Learn more
    This project consisted in designing and implementing a dual core 8051 processor with pipeline capabilities on a FPGA. The processor was designed from scratch, based on the CISC instruction set of the 8051 processor.
    The main components of the design were the translation of the CISC instructions into a set of RISC micro instruction, the jump prediction logic, the pipeline queuing and the management of shared resources between the two cores.
  • 2008

    Facial Recognition

    Areas: Biometric Recognition, Image Processing, Signal Processing,
    Learn more
    This project consisted in identifying people on an image or video stream from a set of known individuals in a database.
    To attain this objective, a Haar cascade detector was used to detect the faces on the image, while an Active Shape Model (ASM) was used to normalize differences in the head position and facial expressions. The features obtained from Principal Components Analysis (PCA) were used to train a Gaussian Mixture Model (GMM) and learn the probability distribution of such features for each individual. These models were then used to identify a face from a set of known individuals in a database.
    The results of this project were presented on the ITBA fair of electronics, where it was chosen as the best R+D project by three of the fair sponsors.