M2 Multimedia Networking

Mihai Mitrea Bernadette Dorizzi Dijana Petrovska.

Courses + Labs.

Objectives: Through this course, students will have the opportunity to: (1) to acquire the fundamental concepts and incremental challenges related to the security of networked multimedia content (protection, integrity, certification, traceability, etc.), (2) to understand the underlying scientific paradigms and (3) to learn about the various practical applications through use cases from the R&D world.

Course content (26 h)
Introduction (2 h)
- Protection, integrity, certification, traceability: a solution for every problem!

Basic Crypto (6 h)
- Encryption
- Authentication

Watermarking and steganography (11 h)
- Definitions, basic properties
- Digital watermarking in the context of information theory:
o spread spectrum watermarking
o on-board information watermarking (continuous and discrete case)
o statistical modelling for attacks (Gaussian case in non-Gaussian, stationary and non-stationary)
- Watermark features by content type: image, audio, video

Biometric security (2 h)
- Definitions, properties
- Applications

Practical work (5 hours)
- Cryptography
- Spatial watermarking methods (LSB - by substitution)
- Watermarking methods in the transformed domain (SS - by spread spectrum).

Probability theory.

Octobre - Novembre - Décembre.

PALAISEAU

Marco Cagnazzo Attilio Fiandrotti.

Courses + Labs.

Objectives: This course aims to give students an overview of modern techniques for compressing audio-visual signals. Based on an introduction to the physiology of perception, this course shows the latest techniques for compressing music (perceptual coders) and images (wavelet transform). The course also provides a sufficiently complete introduction to video coding to enable students to profitably follow the Audiovisual transport course in the 2nd semester.

Course content (25 h)

Perception (2 h)
- Sound, image, video

Coding of audio signals (3 h)
- Perceptual encoders
- Illustration: MP3 and AAC

Wavelet Transform (10 h)
- discrete Wavelets
- Wavelet bases and MRA approaches
- Wavelet families
- Wavelets adapted to the content (x-lets)
- Optimal flow allocation
- Illustration on JPEG 2000. EZW, SPIHT

Video coding (6 h)
- Motion estimation
- Hybrid encoder structure
- H.264 illustration

Practical work (3 hours)
- Inside JPEG 2000.

Knowledge in probability and linear algebra.

K. Sayood, Introduction to Data Compression - 5th Edition, Morgan Kaufmann, 2017.

Octobre - Novembre - Décembre.

PALAISEAU

Michèle Wigger.

Courses + exercises.

Objectives : Provide the foundations in information theory for source coding.

Content of the courses (25 h)

1.Properties of Shannon’s Information Measures

2.Markov Chain, Fundamental Inequalities and Entropy of Stationary Sources

3.Lossless Source Coding

4.Coding Theorem for Lossy Source Coding

5.Distributed Lossless Compression

6.Lossy Compression with Side Information

7.Multiple Description Coding

8.Distributed Lossy Compression.

Good knowledge in probability.

T. M. Cover, J. A. Thomas, Elements of Information Theory, Wiley, 2006.

Octobre - Novembre - Décembre.

PALAISEAU

Véronique Vèque Michèle Wigger Industriels.

Courses + Exercises + Labs.

Define the architectures, protocols and mechanisms that enable IP-based networks to provide the quality of service necessary to support multimedia applications. The QoS mechanisms deployed at the heart of the network but also at the level of access networks as well as the optimization tools and methods used to characterize their performance will also be detailed. Content distribution networks will be presented with an emphasis on models to study their performance and optimize them.

Course content (25 h)

Multimedia applications (1 h)
- Quality of service (QoS)
- Traffic patterns

QoS architecture and protocols (3h)
- Multimedia protocols and signalling: RTP, RTCP, RTSP
- QoS architectures: DiffServ, RSVP, MPLS, VPN

Access networks and QoS (3h)
- QoS Ethernet, Wi-Fi,...

QoS in the network (8 h)
- QoS mechanisms
- Performance and optimization

Adaptive routing, multi-criteria routing (2 h)
- QoS, BGP/BGP-TE routing

Content distribution networks (9 h)
- Principles of content distribution networks
- CDN, CCN, ICN, Peer to Peer, Operators, Client-Server
- Architecture, placement, optimization
- Performance metrics (response time, offloading, availability of services/audiovisual/multimedia flows and data, etc.)
- Performance of distribution networks

Supervised work (2 hours)
- TD scheduling mechanisms

Practical work (3h)
Performance studies and optimization with comparison of different content distribution and media flow systems.

Architecture of TCP/IP networks.

J. Kurose, K. Ross, Computer Networking: A Top-Down Approach, Global Edition, Pearson, 2016.

Octobre - Novembre - Décembre.

GIF-SUR-YVETTE - PALAISEAU

Frédéric Dufaux Michel Kieffer.

Courses + Lab.

Objectives : Present the components of all types of source coders. Present lossless coding, uniform and non-uniform scalar quantization, predictive coding and transform coding. Provide optimal rate allocation when coding multiple sources in parallel.

Course content (18 h)

Lossless coding (6 h)
- Source models, entropy
- Huffman coding
- Coding of Lempel-Ziv
- Arithmetic codes
- Corresponding decoders

Lossy coding (6 h)
- Scalar quantization, Lloyd-Max algorithm
- Optimal bit allocation, Shoam-Gersho algorithm
- Predictive coding

Two-dimensional block transform (6 h)
- Vector and matrix representation of an image
- KLT
- Separable transforms
- Common transforms
- Application to image compression. Coding gain

Practical work (optional)
- Lossless compression: what is behind Winzip?
- Encoding still images: from JPEG to JPEG 2000
- Video coding: from motion-JPEG to H264 and beyond... demystify DivX!.

Basic knowledge in probability and linear algebra.

K. Sayood, Introduction to Data Compression - 5th Edition, Morgan Kaufmann, 2017.

Septembre - Octobre.

GIF-SUR-YVETTE - PALAISEAU

Frédéric Lehmann Marco Cagnazzo.

Courses + Labs.

Objectives: Give the basic tools in probability, signal processing, estimation in order to start source coding courses in good conditions.

Course content (18 h)

Introduction to probability theory (3 h)
- Sampling spaces, events, probability concept
- Discrete or continuous random variable
- Probability laws (distribution function, moments)
- Common laws (binomial, Poisson, Gaussian, exponential...)

Stochastic processes (4 h)
- Poisson Process
- Birth and death processes
- Markov chain with discrete time

Estimation (3 h)
- Maximum likelihood principle
- Estimation of the parameters of a Poisson process and a Markov chain

Digital filtering (8 h)
- Reminders on discrete time linear systems
- Sampling theorem, TF
- Reminders on digital filtering
- Multicadence filtering,
- Polyphase representations
- M-band filter banks.

Practical work (optional) (4h)
- Lab on random signals and estimation
- Lab on filter banks and multi-cadence filtering.

Basic knowledge in probability and linear algebra.

S. M. Kay, Fundamentals of Statistical Signal Processing, Volume I: Estimation Theory, Prentice Hall, 1993.

Septembre - Octobre.

PALAISEAU

José Néto Michel Kieffer.

Courses + Labs.

Objectives: This course aims to provide a basis for optimization. The necessary theoretical foundations will be laid; algorithms and methods will be described with the aim of identifying the situations in which they apply. Some applications may be chosen to illustrate the methods.

Course content (18 h)

Introduction and theory of continuous optimization (4.5h)
- notions of convexity
- types of problems: (convex optimization, linear/quadratic programming, SOCP, SDP,...)
- duality, dual problem, dual function of Lagrange
- optimality conditions (KKT,...)

Continuous optimization algorithms (4.5h)
- descent methods, gradient, Newton, conjugated gradient and quasi-Newton
- notions on advanced methods (inner points,...)
- elements on convergence

Easy problems of combinatorial optimization (4.5h)
- Covering shaft of minimum weight
- Shortest way
- Maximum flow and minimum cut

Difficult combinatorial problems (4.5h)
- Introduction to Complexity (P, NP, NP-complete, etc.)
- Separation method and evaluation
- Polyhedral approaches

Practical work
- use of CLC software (convex optimization)
- programming of simple algorithms
- illustration by applications chosen from: approximation/interpolation/ statistical estimation/geometric problem.

Good knowledge of common linear algebra and function calculation.

S. Boyd, L. Vandenberghe, Convex Optimization, Cambridge University Press, 2004.

Septembre - Octobre.

GIF-SUR-YVETTE - PALAISEAU

Sahar Hoteit.

Courses + Labs.

Objectives: This course recalls the basic elements of the Internet, its architecture and protocols.

Course content (16 h)
Introduction to the Internet (2 hours)
- Introduction on the development and actors of the Internet
- Structure of the Internet
- Internet access

TCP/IP architecture and addressing (3h)
- Addressing
- Addressing and naming

IPV6 (3h)

Main protocols (6h)
- IP Protocol and its tools
- Internet routing
- Main TCP and UDP transport protocols

Main applications (2h)
- HTTP, DNS

Practical work (optional)
- Study of a multimedia protocol: SCTP (3h).

Some basic knowledge in networking.

J. Kurose, K. Ross, Computer Networking: A Top-Down Approach, Global Edition, Pearson, 2016.

Septembre - Octobre.

GIF-SUR-YVETTE - PALAISEAU

KiefferMichel DufauxFrédéric CagnazzoMarco Le FeuvreJean ValenziseGiuseppe CagnazzoMarco HoumaniNesma FiandrottiAttilio.

Bibliographical work + research work + report + defense.

Give the opportunity to students to apply the theoretical notions studied in at least two different courses on a practical problem. Provide an introduction to research problems.

Course outline

Perform an original research work in an academic lab, involving the content of at least two courses presented during the master program.

Octobre - Novembre - Décembre - Janvier - Février - Mars - Avril.

GIF-SUR-YVETTE - PALAISEAU - COURCOURONNES

Mitrea Mihai Dorizzi Bernadette Delpha Claude Le Feuvre Jean.

Courses + Labs.

Objectives: complement the security principles addressed in the first course with advanced methodologies required for content distribution and establish the underlying normative panorama. Interactions and synergies between basic techniques (compression - protection, watermarking - selective encryption - biometrics) will also be presented.

Course content (26 h)

Techniques to optimize the performance of watermarking systems (6 h)
Quantity of information inserted: On-board information techniques
QIM and mQIM (Quantization Index Modulation)
Other on-board information techniques

Securing the MPEG-4 AVC and HEVC compressed flow (2 h)
Watermiarking in the compressed domain
Selective encryption

Collection, extraction and analysis of evidence (forensics) (2 h)
Basic methods
Countermeasure method (anti-forensics)

Traceability of content (3 h)
Definitions, properties, applications
Towards a theoretical model

Quantum cryptography concepts (2 h)
Quantum information and security
Quantum key distribution

Normative framework (3 h)
Standards for broadcast protection (DVB)
Standards for OTT protection (CENC, ISMA)
MPEG standards for content management
Standards for protection in digital cinematography (DCI)

Examples of security methods through use cases from the R&D world (8 h)
Real-time and on-demand service protection
IPR protection for film and TV/TV-3D content; joint protection techniques
Authentication of video surveillance content
Tracking of online content
Biometrics.

Multimedia content security: basic knowledge.

Janvier - Février - Mars.

PALAISEAU

Michel Kieffer.

Courses + Labs.

Objectives: The purpose of this course is to present a number of tools for modelling and estimating the characteristics of flows and demands for audiovisual content. It then presents techniques from control theory (fluid modeling and stochastic control) for the study of end-to-end regulatory mechanisms for content delivered over a network.

Course content (25 h)

Fundamentals of prediction and learning (9 h)
Prediction of flows and demands and learning
to identify patterns that can be used for:
adaptation and optimization of flow processing, distribution and support
feed management and control
analysis and correlation methods, in particular on performance indicators and monitoring

Fluid modeling and network stability (8 h)
Modeling of source control mechanisms, congestion control mechanisms,
Criteria for local and global stability of regulatory mechanisms,
Impact of processing and transmission times on stability

Micro level (nodes, flows), distributed, view located with local information (8 h)
Stochastic control and Markovian decision-making processes
Concepts of states, automata, utility function, reward, policies
Iteration on values or policies to obtain optimal policies
Learning strategy for scaling up
Multi-layered CDM - Markovian processes in interactions
Illustration: inter-layer optimization of scalable video streaming streams.

Basic course in probability Source coding course.

R. S. Sutton, A. G. Barto, Reinforcement Learning: An Introduction, The MIT Press, 2018.

Janvier - Février - Mars.

GIF-SUR-YVETTE - PALAISEAU

Attilio Fiandrotti Giuseppe Valenzise.

Courses + Labs.

Objectives: Deep learning based approaches have recently outperformed existing techniques in a number of signal processing and multimedia applications. The availability of large sets of annotated data, powerful yet affordable GPUs and novel optimization techniques are at the core of such revolution. The present course will cover topics such as convolutional, recurrent and adversarial neural networks, the relative training procedures and the existing frameworks for deep learning. The course will focus in particular on the application of such technologies to problems such as as image classification, object segmentation, text reading and video compression.

Content of the course (24 h)

Introduction (1 h)
Machine learning taxonomy and basics
Example applications

Feed-forward Neural Networks (3 h)
The single layer perceptron
The single layer sigmoid network
The error function
Stochastic gradient descent

Multilayer Neural Networks (2 h)
Multi layer sigmoid networks
Backpropagation
Fully connected networks for character recognition: LeNet300

Convolutional Neural Networks (3 h)
Parameter sharing
CNNs for character recognition: LeNet5
Deep learning frameworks: PyTorch

Deep neural networks (3h)
The vanishing gradient problem
The ReLU activation
Regularization
Optimization for deep architectures (AdaGrad, drop out, batch norm, etc.)
Noteworthy architectures: AlexNet, ResNet, etc.

Labs (12h).

Course in probability.

I. Goodfellow, Y. Bengio, A. Courville, Deep Learning, The MIT Press, 2016.

Janvier - Février - Mars.

PALAISEAU

Jean Le Feuvre.

Courses + Labs.

Objectives : The aim of this course is to learn the fundamental concepts of audiovisual content transport, understand how these concepts are used in existing transport architectures, protocols and formats and to experiment with these protocols and standards to open up to emerging techniques and ongoing research in the field.

Course content (25 h)

General principles of audio-visual transport (3 h)
Types of transport chains: broadcast, multicast, unicast, P2P
Real-time constraints, quality of service
Media Synchronization
Optimization of network resources (packetization, multiplexing/interleaving)

Real-time audio-visual (3h)
Real-time transport (RTP/RTCP)
Signalling/session control techniques 
Service deployments (RTSP, SIP)

Digital TV (3h)
MPEG-2 Transport Stream
Associated standards (DVB, ATSC, ISDB)
Service deployments (DVB network architecture, IPTV networks)

Broadcast on HTTP (4h)
File formats
Streaming techniques via HTTP 
Adaptation in flow and associated protocols (MPEG-DASH, HLS)

Advanced transport applications (6 h)
Connected television, hybrid multi-channel and multi-path broadcasting, multi-terminal synchronization
Audio-visual in web browsers
Dissemination via peer-to-peer networks

Supervised work (6 hours)
Case study for digital TV
RTP Streaming
MP4 and MPEG-DASH file format.

Networking: protocols (IP, TCP, UDP, HTTP, HTTPS), Internet architecture (DNS, unicast/multicast, CDN, P2P) Coding: video compression, audio compression.

Janvier - Février - Mars.

PALAISEAU

Marco Cagnazzo.

Courses + Labs.

Objectives: This course complements the Compression of multimedia content course, by introducing advanced coding techniques. Mathematical models of quality are illustrated, as well as vector quantification techniques. Then, the most important standards in video coding are analyzed in terms of mathematical compression tools. Mathematical coding models are used to introduce quality optimization and rate control techniques. Techniques to code new formats (3D, HDR) and to better adapt to network characteristics are illustrated at the end of the course.

Course content (25 hours of course)

Quality evaluation (2 h)
- Perceptual models (audio, image and video)
- Quality metrics

Vector quantization (2 h)
- Definition
- Lloyd-Max, LBG algorithm
- Perf study (Bennett formula), coding gain
- Implementation

Video encoders (4 h)
- Classic encoders: MPEG-2
- Modern encoder: H.264, HEVC

Control techniques and optimization for distribution (2 h)
- Rate-distortion optimization
- Rate control

Advanced coding techniques (9h)
New formats :
- 3D coding, multi-view (2h)
- HDR (1h)
Coding and transmission:
- Multiple description coding (2h)
- Scalable coding (2h)
- Distributed coding (2h)

Practical work (6 hours)
- 3D video
- QV.

First course in source coding.

Janvier - Février - Mars.

PALAISEAU

Michel Kieffer Mihai Mitrea Marco Cagnazzo Sahar Hoteit.

Research within a lab Bibliography Research work Report Defense.

Get a first research experience in a industrial or academic lab.

Perform an original research work in a lab.

Avril - Mai - Juin - Juillet.

GIF-SUR-YVETTE - PALAISEAU - COURCOURONNES

Depends on the proposed seminars.

Courses.

Objectives:
present more advanced research topics linked to multimedia networking in general. Present some open research directions. Help students defining their own research topics. Three seminars have to be followed by students, with an evaluation on two of them.

Course content:
Choice among
- Quantum Information
- Joint source/protocol/channel coding and decoding
- Advanced 3D coding techniques
- Medical applications in distributed environments
- Optimization of the diffusion technique (uni/multi/broad cast)
- Network coding for multimedia
- Biometry
- ….

Novembre - Décembre - Janvier - Février - Mars.

GIF-SUR-YVETTE - PALAISEAU