Next:
List of Figures
Up:
Thesis
Previous:
Remerciements
Index
Contents
Remerciements
Contents
List of Figures
List of Tables
Publications
Acronyms
Glossary
11
Évaluation automatique de la qualité des flux multimédias en temps réel : une approche par réseaux de neurones
Introduction
Les motivations
Les contributions de cette thèse
Vue d'ensemble de la dissertation
Descriptions de notre nouvelle méthode en général
Vue d'ensemble de la méthode
Essais subjectifs de la qualité
Calcul de MOS et analyse statistique
Comparaison entre ANN et RNN
Choix de paramètres
La mesure de la qualité de la parole en temps réel
Paramètres affectant la qualité de la parole
Une base de données de MOS pour différentes langues
Évaluation de la qualité de la parole par le RN
Performance d'autres mesures de la qualité de la parole
Évaluation de la qualité de la vidéo
Les paramètres influant la qualité de la vidéo
Résultats
Les performances des autres mesures de la qualité
Études des effets de paramètres sur la qualité
Impact des paramètres sur la qualité de la parole
Impact des paramètres sur la qualité de la vidéo
Un nouveau mécanisme de contrôle de débit
TFRC basé sur une équation (EB-TFRC)
Notre protocole de contrôle de débit
Résultats de simulation
Une nouvelle méthode pour la prévision du trafic
Notre méthode proposée
Les résultats et les évaluations expérimentales du nouveau modèle
Nouveaux algorithmes d'apprentissage pour les réseaux de neurones aléatoires
La méthode de
Levenberg-Marquardt
pour RNN
LM avec
adaptive momentum
pour RNN
Évaluation des performances des algorithmes proposés
Conclusions générales
Sommaire des contributions principales
Extensions possibles
English Annex: Automatic Evaluation of Real-Time Multimedia Quality: a Neural Network Approach
Introduction
Motivations
Multimedia Quality Assessment
Impact of the Quality-Affecting Parameters on Multimedia Quality
Rate Control Protocols
Traffic Prediction
Neural Network Learning
The Contributions of this Dissertation
Overview of the Dissertation
Automatic Real-Time Multimedia Quality Evaluation
State of the Art
Objective Speech Quality Measures
Bark Spectral Distortion (BSD)
Enhanced Modified BSD (EMBSD)
Perceptual Speech Quality Measure (PSQM)
PSQM+
Measuring Normalizing Blocks (MNB)
Perceptual Analysis Measurement System (PAMS)
ITU E-model
Objective Video Quality Techniques
ITS Video Quality Measure
EPFL Objective Video Quality Measures
Other Works in Objective Video Quality Measures
Multimedia Transport Protocols
Challenges of Multimedia Transmission
Working Around Loss
Video Compression and Standard Codecs
Spatial or Block Coding
Temporal Coding
Standard Video Codecs
Audio Compression and Codecs
Waveform Coding
Source Codecs
Hybrid Codecs
Standard Audio Codecs
Neural Networks
Artificial Neural Networks (ANN)
Random Neural Networks (RNN)
Descriptions of Our New Method in General
Introduction
Overview of the Method
More than one Media Type
Selecting the Parameters' Values
Subjective Quality Tests
Source Signal for Audiovisual Tests
Number of Subjects and their Selection
Instructions for Assessors (Subjects)
The Test Sessions
Subjective Quality Test Methods
Comparison of the Methods
MOS Calculation and Statistical Analysis
On the Use of the Neural Networks
Comparison between ANN and RNN
Parameter Selection
Expected Use of Our Method
Possible Uses For Speech and Audio
Possible Uses for Video Applications
Operation Mode
Conclusions
Measuring Speech and Video Quality, and Applications
Measuring Speech Quality in Real-Time
Introduction
Measuring Network Parameters in a Testbed
Speech-Quality-Affecting Parameters
Other Effects
Our Method and Session Types
A Mean Opinion Score (MOS) Database for Different Languages
Assessment of Speech Quality by Neural Networks
Performance of other Speech Quality Measures
Conclusions
Measuring Video Quality
Introduction
Simulator Description
H263 encoder
H263 decoder
Network Transport Simulation
The Quality-Affecting Parameters
Subjective Quality Test and MOS Experiment
Results
Training the NN
How Well does the NN Perform?
Performance of other Video Quality Measures
A Demonstration Application
Conclusions
Study of Parameters Effects on the Quality
Introduction
Related Works
Parameters Impact on Speech Quality
Loss Rate (LR)
Consecutive Lost Packets (CLP)
Packetization Interval (PI)
Speech Codec
Parameters Impact on Video Quality
Bit Rate (BR)
Frame Rate (FR)
Loss Rate (LR)
The number of Consecutively Lost Packet (CLP)
Intra-to-Inter Ratio (RA)
Conclusions
A New Rate Control Mechanism
Introduction
Related Works
Limitation of RTCP in Rate Control
Equation-based TFRC
Our Proposed Rate Control Protocol
The Possible Controlling Parameters
Simulation Results
Speech Case
Video Case
Discussion
Conclusions
On the Neural Networks Tools
Using Neural Networks for Traffic Prediction: a New Method
Introduction
Our Method
Experimental Results and Evaluation of the New Model
Real Traces for Training
Training Database Description
Experimental Tests to Identify the Best Length of Each Window
Performance of the NN to Predict Traffic
More Than One Step in the Future
Conditions to Retrain the NN
Traffic Flow Type
Comparison with Other Models
Possible Uses of our Model
Conclusion and Discussion
New Random Neural Network Training Algorithms
Introduction
Gradient Descent Training Algorithm for RNN
New Levenberg-Marquardt Training Algorithms for RNN
Analytical Formulation of the Algorithm
Different Variants of the LM Training Algorithm
New LM with Adaptive Momentum for RNN
The Algorithm with Adaptive Momentum
Performance Evaluation of the Proposed Algorithms
and
Parameters for AM-LM
Algorithms' Performance Comparison
Testing the Success Rate and the Performance
Learning Performance in the Video Quality Problem
Conclusions
General Conclusions of this Dissertation
Summary of the Main Contributions
Possible Extensions
Appendix
Bibliography
Index
About this document ...
Samir Mohamed 2003-01-08