We propose a joint source-channel coding scheme, developed for video sequences, which consists of a vector quantization based on lattice constellations and a linear labelling minimizing simultaneously the source and the channel distortion. The linear labelling has already been proved to minimize the channel distortion on binary symmetric channels and the linear transforms based on lattice constellations of maximum diversity to minimize, at the same time, the distortion of Gaussian sources. We study the dependencies between the wavelets coefficients of a $t+2D$ video decomposition in order to efficiently exploit the linear transforms developed for Gaussian sources. As the source distribution of the subbands is not Gaussian we present the necessary modifications in order to obtain a robust coding scheme. We propose a stochastic model to capture the dependencies between the wavelets coefficients and we use it to build an optimal mean square predictor for missing coefficients. We present two applications of this predictor on the transmission over packet networks: a quality enhancement technique for resolution scalable video bitstreams and an error concealment method. We develop a robust joint source-channel coding scheme for transmission of video sequences over a Gaussian channel using uncoded and coded index assignment via Reed-Muller codes. We investigate the conditions requiring the use of a coded index assignment and we prove its superiority compared to an unstructured vector quantizer. For a transmission over a flat-Rayleigh channel, we develop a robust coding scheme using our vector quantizer followed by a rotation matrix.
Authors
- Bibliographic Reference
- Georgia Feideropoulou. Codage Conjoint Source-Canal des Sources Vidéo. Traitement du signal et de l'image [eess.SP]. Télécom ParisTech, 2005. Français. ⟨NNT : ⟩. ⟨pastel-00001294⟩
- HAL Collection
- ['Institut Mines Télécom', 'PASTEL - ParisTech', 'ParisTech', 'CNRS - Centre national de la recherche scientifique', 'Ecole Nationale Supérieure des Télécommunications', 'Télécom Paris', "Laboratoire Traitement et Communication de l'Information", 'composantes instituts telecom']
- HAL Identifier
- 503439
- Institution
- Télécom ParisTech
- Laboratory
- Laboratoire Traitement et Communication de l'Information
- Published in
- France
Table of Contents
- Georgia Feideropoulou 2
- Abstract 4
- Résumé 6
- Remerciements 8
- Table des matières 10
- Table des figures 14
- Liste des tableaux 18
- List of Notations 22
- Statistical models for wavelet coefficients 22
- Spatio-temporal modelling of the wavelet coefficients in a scheme 22
- Overview of Joint Source-Channel coding schemes 22
- Joint Source-Channel coding based on linear labelling and rotated constellations 22
- Joint Source-Channel coding of a video on a Gaussian channel 23
- Joint source-channel coding of a video on a flat-Rayleigh fading channel 23
- List of abbreviations 24
- Résumé de la thèse 26
- Introduction-Motivation 26
- Codage conjoint source-canal de sources gaussiennes 26
- Constellations à diversité maximale 27
- Codage conjoint source-canal de séquences vidéo 29
- Modèle Statistique 29
- Applications du modèle statistique 31
- Construction du dictionnaire de source pour des séquences vidéo 32
- Premier résultats sur des canaux Gaussien et Rayleigh 34
- Algorithme dallocation de débit 35
- Application dalgorithme dallocation du débit 36
- Schéma du codage avec un étiquetage linéaire codé sur un canal gaussien 37
- Comparaison avec un quantificateur vectoriel non-structuré 39
- Schéma de codage avec matrice de rotation sur un canal Rayleigh 40
- Conclusions 42
- Chapitre 1 46
- Statistical models for wavelet coefficients 46
- 1.1 Introduction 46
- 1.2 Overview of statistical wavelet models for still images 47
- 1.2.1 Interscale Models 47
- 1.2.2 Intrascale Models 48
- 1.2.3 Composite Models 48
- 1.3 Simoncellis Joint Statistical Model 48
- 1.4 Introduction to the scheme of decompositon of a video sequence 50
- 1.5 Overview of statistical wavelet models for video 52
- 1.5.1 Marginal distribution of the spatio-temporal wavelets coef- ficients 52
- 1.5.2 Extension of models for still images to the video domain 54
- 1.6 Conclusion 56
- Chapitre 2 58
- Spatio-temporal modelling of the wavelet coefficients in a scheme 58
- 2.1 Introduction 58
- 2.2 Conditional histograms of wavelet coefficients in a scheme 58
- 2.3 Double Stochastic Model 59
- 2.4 Model Estimation 64
- 2.4.1 65
- 2.4.2 66
- 2.4.3 67
- 2.5 Illustration Examples 68
- 2.6 Conclusion 70
- Chapitre 3 72
- Application of the statistical model to error concealment and quality enhancement of video 72
- 3.1 Introduction 72
- 3.2 Prediction Method 73
- 3.3 Model-Based Quality Enhancement of Scalable Vi- deo 74
- 3.4 Error concealment in the Spatio-temporal wavelet domain 75
- 3.5 Error concealment of scalable bitstreams 80
- 3.6 Conclusion 82
- Chapitre 4 84
- Overview of Joint Source-Channel coding schemes 84
- 4.1 Introduction 84
- 4.2 Problem Statement 85
- 4.2.1 Vector Quantization 86
- 4.2.2 Index assignment IA 87
- 4.2.3 Channel coding 92
- 4.2.4 Channel decoding 93
- 4.3 Source-optimized channel coding 95
- 4.4 Channel-optimized source coding 100
- 4.5 Other combined optimizations 102
- 4.6 Conclusion 103
- Chapitre 5 104
- Joint Source-Channel coding based on linear labelling and rotated constellations 104
- 5.1 Introduction 104
- 5.2 Linear Labelling and Joint Source-Channel coding 105
- 5.2.1 Linear labelling to minimize channel distortion 105
- 5.2.2 Minimisation of source distortion-case of Gaussian sources 107
- 5.3 Maximum component diversity constellation 108
- 5.4 Construction of rotated -lattices of dimension and mixture constructions 109
- 5.4.1 Mixture constructions of rotated -lattices 110
- 5.5 Rotated lattices from cyclotomic fields where is a power of 111
- 5.6 Sphere Decoder 112
- 5.7 Performances of rotated BPSK over a Rayleigh fa- ding channel 114
- 5.8 Conclusion 115
- Chapitre 6 116
- Joint source-channel coding of a video on a Gaussian channel 116
- 6.1 Introduction 116
- 6.2 Bit allocation algorithm 117
- 6.2.1 A general method of bit allocation for spatio-temporal sub- bands 117
- 6.2.2 A robust bit allocation algorithm 119
- 6.3 Coding algorithm 120
- 6.3.1 Source codebook construction 121
- 6.3.2 Scaling the lattice constellation to the source dynamics 123
- 6.4 Application of joint source-channel coding scheme of a video over a Gaussian channel 126
- 6.4.1 A First Attempt 126
- 6.4.2 Application of the bit allocation algorithm 129
- 6.5 Calculation of the end-to-end distortion in the noisy system 133
- 6.6 Vector quantization by linear mapping of a block code 134
- 6.6.1 Index assignment using 135
- 6.6.2 Overall results when is used for index assignment 139
- 6.7 Structured vs. unstructured codebook with index assignment 140
- 6.8 Comparison with a MC-EZBC protected with rate punctured convolutional codes 145
- 6.9 Conclusion 148
- Chapitre 7 150
- Joint source-channel coding of a video on a flat-Rayleigh fading channel 150
- 7.1 Introduction 150
- 7.2 Linear Labelling and Puntured Convolutional Codes on a flat-Rayleigh fading Channel 151
- 7.3 Joint source-channel coding using rotations prior to transmission over a flat-Rayleigh channel 153
- 7.4 Comparison with the MC-EZBC protected by rate punctured convolutional codes over a flat-Rayleigh channel 156
- 7.5 Conclusion 158
- Conclusions and Perspectives 164
- Annexe A 168
- Basic definitions in Algebraic Number theory 168
- A.1 General definitions 168
- A.1.1 Properties of Ideals 171
- A.2 Ideal Lattices 171
- A.2.1 Construction of rotated - lattices with full diversity by Ideal lattices 172
- A.2.2 Cyclotomic fields 173
- Annexe B 174
- Vectorial extension of the double stochastic model 174
- Annexe C 176
- Rotation Matrices 176
- Bibliographie 178
- List of publications 185