project: Implementation of a novel on-board hyperspectral data
In 2007, the European Space Agency (ESA) granted 150,000 € for supporting a 1-year feasibility study on the application of my Ph.D. thesis to the onboard compression of satellite images. This achievement was the result of a 6-month personal initiative to bring together industrial and academic partners around the applications of my Ph.D. thesis. The project was carried out in the context of an Innovation Triangle Initiative (ITI), supported by ESA with LuxSpace, OHB System AG and Supélec as partners. The main goal of the project was to develop and demonstrate the feasibility and use (in a laboratory environment) of a demonstrator for an on-board satellite image compression system based on the concept of "Optimal Transform Codes" (OTC) developed during my Ph.D. thesis.
Independent component analysis (ICA) and transform coding for image/video compression
From 2002 to 2005 I worked as a Research Assistant in the "Information, Multimodality & Signal" (IMS) research group at Supélec (Metz, France). I was interested in independent component analysis (ICA) and transform coding for signal (image/video/speech) compression. In this position I first proposed a new point of view in transform coding: the problem of finding the optimal 1-D linear block transform may be viewed as a modified ICA problem. This result applies without the presumption of Gaussianity. By adopting this new viewpoint, I then derived two new ICA-based algorithms, called GCGsup and ICAorth, for computing the optimal 1-D linear transform and the optimal 1-D orthogonal transform, respectively. Experimental results showed that the new transforms can achieve better visual image quality than the classical discrete cosine transform (DCT) used in JPEG and MPEG. The new transforms proved also efficient for the compression of multicomponent images such as multispectral and hyperspectral satellite images. Performance comparisons with the classical Karhunen-Loève transform (KLT) showed that the transforms returned by GCGsup and ICAorth can achieve better spectral redundancy reduction.
Emergence of Simple-Cell Receptive Field Properties by Learning Optimal Variable-Rate Transform Codes for Natural Images
The receptive fields of simple cells in mammalian primary visual cortex can be characterized as being spatially localized, oriented and bandpass, comparable to the basis functions of wavelet transforms. A fundamental problem in vision research is to determine why the receptive fields and response properties of visual neurons are as they are. One approach to understanding such response properties has been to consider their relationship to the statistical structure of natural images in terms of efficient coding. Models for learning efficient codes for natural images, such as sparse coding or ICA, predict the localized, oriented, and bandpass characteristics of simple cells. In this work, I propose a new approach to understanding the response properties of simple cells, which has its roots in the framework of variable-rate transform coding. Recently, we proposed a new viewpoint in variable-rate transform coding. Under the high resolution hypothesis, we showed that the optimal variable-rate transform code for a signal to be encoded is the solution of a modified independent component analysis (ICA) problem. This result applies without the presumption of Gaussianity or orthogonality. By adopting this new viewpoint, we derived a new algorithm, called GCGsup, for learning optimal variable-rate transform codes from a set of signals (e.g., images and sounds) to be encoded. In this work, we show that optimal variable-rate transform codes for natural images exhibit a complete family of localized, oriented, bandpass receptive fields, similar to those found in the primary visual cortex.
Acoustic mine imaging and sonar
From July 2000 to November 2001 I worked as a R&D Sonar Engineer in the General Sonar Studies Group of Thales Underwater Systems Pty Ltd, Sydney, Australia. I worked on a 3-D acoustic mine imaging project, the purpose of which was to build an acoustical system capable of overcoming the limits of an optical camera operating in turbid water. In the context of this project, I developed a computer model for the signal processing chain of the acoustical camera. My research work enabled the definition of an optimal array of sensor, which was essential for obtaining the best image quality.
Automated breast cancer diagnosis
From March 2000 to November 2001 I worked in the Laboratory of Experimental Oncology of the French Research Center for Medical Research, Marseille, France as a M.S. trainee. I worked at clinical site on an interdisciplinary team of image processing researchers, biologists, and medical practitioners to investigate new methods for automated breast cancer diagnosis. In this position I first designed and implemented an algorithm for the segmentation of histological structures in microscopic images using mathematical morphology operations. Then I proposed a supervised learning technique for predicting the histologic grade --- in the Scarff-Bloom-Richardson system --- of a patient's cancerous breast tumor. A good match between the grades returned by my predictor and those obtained from visual inspection by a medical practitioner was observed.
Management of Quality of Service (QoS) in digital T.V. operations covered by the MPEG-2 standard
From July 1999 to October 1999 I worked in the Design and Test of Equipment Laboratory of the French Research Center in Broadcasting and Radiocommunications (TDF) as an Engineer trainee. I worked on a project related to the management of Quality of Service (QoS) in digital T.V. operations (satellite, cable and terrestrial networks) covered by the MPEG-2 standard. In the context of this project, I designed and built an experimental set-up for assessing the performance of a video quality evaluation system.
Simulation of a fluid flow
During summer 1996 I worked in the R&D department of Z.F., Saarbrücken, Germany as a Summer Intern. I performed some simulations (using Matlab and AutoCAD) for the optimization of the electro-hydraulic brain of an automatic gearbox.
Last updated: December 2010Michel Narozny - Copyright 2006-2010