Professional Documents
Culture Documents
Objectives:
Companies that develop and/or make use of image processing methods are extremely numerous
and varied in terms of activities as in terms of size. To illustrate our statement, here is a non-
exhaustive list of applications domains: medical and biological purposes, security and surveillance,
multimedia, photo and video, robotic, biometrics and remote sensing. Moreover, if there exist lots
of small and medium-sized businesses, including start-ups, (for example Timeat, Realviz, Theralys
...), many big companies can be easily identified thanks to their activities that require image
technologies (like Philips, Canon, EADS, Thalès …). But it is often unknown that others great firms
without any visible link with image processing, as automobile or optical industries (Renault, Essilor
…), also work with images for some of their products or services. Have a look to a non exhaustive
list here: http://www.int-evry.fr/citi/TAI/debouches.php.
The program « Image Processing and its Applications » is intended for engineer students who
wish for acquiring knowledge and skills in leading-edge image techniques. Those ones will be
illustrated through three rising-up technologies: Biometrics, Remote Sensing and Virtual Reality.
Objectives:
Organisation:
This program is part of the Advanced Engineering Cycle which covers the 8th and 9th semesters of
the Telecom INT curriculum. Its building bricks consist of 6 independent courses (each representing
45h lectures & labs and 90h homework) which define a complete graduate programme in
multimedia engineering.
The Image Processing and its Applications program is organized in the following Teaching Units:
S8 : - IMA4508 : Image, video & 3D graphics compression
- IMA4509: Visual content analysis
S9 : - IMA5511 : Pattern Recognition and Biometrics
- IMA5512 : Remote Sensing
- IMA5513 : Statistical image processing
- IMA5514 : Multimedia interacting and Virtual reality
A final project (IMA5515), representing 225h homework, will stand during the whole 9 th semester
and will be carried out by small group of one to three students.
Assessment:
Two-student group project (P) (45h) linked to real industrial applications or to national/European
research projects with oral defence (D).
Final mark = Average (P, D)
Objectives:
Keywords:
Image compression, mono / multi resolution coding techniques, transform-based approach,
predictive methods, JPEG/MPEG standards, scalability, progressive transmission, technological
convergence.
Prerequisites:
Course outline:
• New challenges for multimedia compression: digital terrestrial TV, High-Definition TV, TV
over ADSL, scalability and technological convergence
• Generic principles of image compression techniques
• Decorrelation techniques, predictive approaches, transform-based and hybrid methods
• Quantization techniques
• Binary coding: arithmetic coding, error resilient tools
• Multiresolution image coding: wavelet-based approaches
• Scalable compression techniques
• Fractal-based image coding
• The JPEG standards: from JPEG to MotionJPEG
• The MPEG standards: from MPEG-1 to MPEG-4
• Compression for film distribution over the Internet
Faculty:
From INT
- Dr. Marius PREDA
- Prof. Françoise PRETEUX
Guest lecturers
- Dr. Gérard MOZELLE (Thomson)
Assessment:
The assessment pattern relies on supervised personal work, and consist of 3 short presentations
(S1-S3), equivalent to 15h homework, and a 30h group project (2-4 student groups) resulting into a
written report (E1). The following topics:
- visual feature extraction
- variational / morphological / stochastic / statistical image segmentation
- dynamic segmentation and object tracking
- deformation analysis and shape variability modelling
will be used as building blocks for competence acquisition. The proposed subjects will be linked to
industrial applications or to national/European research projects, and will focus on implementing,
following a problem analysis step, the functionalities and technologies presented during the course.
Final grade = Average (E1, Average (S1, S2, S3))
In 2007:
1st session = Average (E1, Average (S1, S2, S3)) (C1)
2nd session = 1 written exam (C2)
Final grade = Max (C1, Average (C1, C2))
Objectives:
- To master the core techniques for low-level visual content (2D/3D still images and videos)
analysis, as a preliminary structuring step towards interpretation and content-based access.
- To understand the related technological and economical challenges, and to gain insight into
emerging application issues.
- To turn into practice computer vision applications by means of visual content analysis (human
motion, object detection,…).
Keywords:
Visual feature extraction, segmentation & grouping, motion estimation & tracking, shape analysis.
Prerequisites:
None
Course outline:
• Visual content analysis: economical and industrial issues, technological challenges and new
services in the Information and Communication Society
• 2D/3D modelling :
- Low-level and high-level attributes
- Geometric, deterministic, stochastic and fuzzy approaches
Faculty:
From INT
- Pr. Françoise PRETEUX
- Dr. Catalin FETITA
- Dr. Titus ZAHARIA
Guest lecturers
- Representatives of industry (DGA, Thalès, Philips)
Organisation: Class work: 45h Home work: 45h Total workload: 90h
Lectures: 15h Tutorial: 12h Labs: 18h
The course is structured in two parts : the first, more theoretical, is focused on basic Pattern
Recognition tools that are required to understand how identity verification is performed on the basis
of a person’s biometric traits; the second is focused on the application of such tools to the field of
identity verification.
Assessment:
Validation is based on 3 evaluated Lab sessions (Lab1, Lab2 and Lab3) and an oral exam (O).
Final Mark = 1/3 [Average (Lab1, Lab2, Lab3) + 2*O]
Objectives:
- Master the tools for pattern recognition and data classification
- Knowledge of the specific techniques of the different biometric modalities in terms of the
general tool adaptation to each of them
- Be able to implement a biometric system of identity verification
Keywords:
Biometrics, face recognition, on-line signature verification, iris recognition, speaker verification
Prerequisites:
Notions of Statistics and Probability Theory (Course “Introduction aux statistiques”, ST21)
Course outline:
First Part: Basics of Pattern Recognition
Introduction
Bayes Classifier
The Linear Model
The K Nearest Neighbor Rule
Hidden Markov Models
Principal Component Analysis, Discriminant Analysis
Multilayer Perceptrons
Kohonen Feature Maps
Instructors:
- Prof. Bernadette Dorizzi
- Dr. Dijana Petrovska
- Dr. Sonia Salicetti
Organisation: Class work: 45h Home work: 45h Total workload: 90h
Lectures: 30h Tutorial: 0h Lab: 15h
Some lectures will be provided by an industry engineer. The labs sessions are carried out by student
pairs during 3 hours and include the taking in hand of software like TeraVue and the ORFEO
Toolbox from CNES.
Assessment:
Final grades in this class will be based on lab work reports.
Final mark = Average (Lab works).
Objectives:
- Be able to implement and optimize the segmentation and classification technologies in the
context of remote sensing
- Master the characteristics of the different types of imaging sensors used in remote sensing
applications
- Knowledge of stakes and applications of remote sensing
Keywords:
Satellite and airborne images, SAR, hyper-spectral images, data fusion
Prerequisites:
Bayesian classification, notions of statistics (UV ST21), low-level image analysis
Course outline:
- Basis of satellite imaging: optical, radar, synthetic aperture radar sensors …
o Applications: land-cover classification (SPOT), oil slick detection (SAR), extraction of
vegetation indicates
- Multi- and hyper-spectral imaging: data reduction (ICA, wavelets), spectral signature, texture
characterisation and segmentation
o Applications: galaxy detection, agricultural and environmental mapping (CASI images)
- Change detection: detection theory, vector-machine support classification
o Applications: glacier development following, devastated zone mapping after a disaster
- Stereovision : epipolar geometry
o Applications in urban area: 3D model construction of a urban scene
- Introduction to geographical information systems (GIS)
Instructors:
- EADS
- Thalès
- Dr A. Joannic-Chardin
- Dr Grégoire Mercier
- Pr Jean-Marie Nicolas
- Dr Michel Roux
- Dr Florence Tupin
Organisation: Class work: 45h Home work: 45h Total workload: 90h
Lectures: 24h Tutorial: 0h Lab: 21h
The labs sessions are carried out by student pairs during 3 hours.
Assessment:
Final grades in this class will be based on lab work reports.
Final mark = Average (Lab works).
Objectives:
- Master the main probabilistic modelling and newest statistical methods used in the
treatment of data masses
- Master their applications in data classification and image segmentation
Keywords:
Hidden Markov models, pairwise and triplet Markov models, Bayesian segmentation, theory of
evidence, sensor fusion, multiresolution, unsupervised segmentation
Prerequisites:
Notions of Statistics and Probability Theory (Course “Introduction aux statistiques”, ST21)
Course outline:
- Hidden Markov models (Gibbs random fields, Markov random chains, Markov random trees)
- Unsupervised statistical segmentation:
o Bayesian segmentations (MAP and MPM)
o learning methods (EM, SEM, ICE)
- Unsupervised statistical fuzzy segmentation
- Estimation of generalised mixing
- Theory of evidence
o Evidential hidden Markov models
o Dempster-Shafer fusion and multi-sensor segmentation
- Pairwise and triplet Markov models
- Markov models and multiresolution (renormalisation group, multigrid algorithm,
multiresolution analysis, models on hierarchical graphs)
- Markov random trees and wavelets
Some lectures could be provided in english, while some others will be immediately followed by lab
work on computers. All tutorials sessions include lab work by student pairs.
Assessment:
Final grades in this class will be based on lab work reports.
Final mark = Average (Lab works).
Objectives:
This teaching unit will provide to future engineer the necessary skills for being able to:
- Represent multimedia signals: image, sound and video
- Analyse speech
- Analyse fixed and moving images
- Synthesize speech and images
- Animate virtual objects
- Interact in a virtual share environment (in network)
- Understand augmented reality techniques
Keywords:
Coding, speech recognition, image synthesis, virtual reality, content based indexation
Prerequisites:
Signal processing (frequency representation…) and C/C++ programming basis.
Course outlines:
- Sound and speech coding.
- Image coding: reminders and comparison between coding norms of images and videos
- Speech recognition.
- Image indexation.
- Motion capture.
- Speech synthesis from text.
- Rendering from multiple views.
- Image synthesis: 3D modelling, Gouraud and Phong shading, ray tracing.
- 3D animation: key points, direct and inverse cinematic.
- Languages and norms for Virtual reality: VRML, WEB 3D and Java3D.
- Virtual collaborative environments and their applications.
- Augmented reality.
Instructors:
- André Bideau,
- Dr Jérôme Boudy,
- Dr Patrick Horain,
- Dr William Navarro (Nortel Networks),
- Pr Yannick Rémion (URCA)…
Organisation: Face to face: 27h Personal work: 198h Total workload: 225h
TAI major’s project is done during the whole semester 9. Each student must make a project with
two or three other students. Planning time slots are dedicated to the project. Meeting with project
manager take place about every other week.
Three types of projects are proposed to students: Experimentation projects, study projects for
companies or administrations, research projects.
Assessment:
The validation of this project is based on the writing of a report (R) and an oral presentation (P).
Samples of subjects:
-
-
-
Faculty:
All lecturers of TAI major