Quality enhancement and assessment of stereoscopic 3D content based on structural information

Sujets de thèse 2014

Intitulé de la thèse
Quality enhancement and assessment of stereoscopic 3D content based on structural information
Publication du sujet sur le site de l’ABG : OUI
Nature du financement : Financement institutionnel, Contrat Doctoral, Financement régional, Contrats université sur projets,)
Spécialité de doctorat : Image Signal et Automatique

Lieu de travail
France/Norvège
Date Limite de candidature : 15/05/2014
Laboratoire d’accueil : XLIM/SIC

Présentation de l’équipe de recherche
Les activités de recherche de l’équipe ICONES sont organisées autour du traitement, de la caractérisation et de l’analyse de signaux et images multi sources et multi composantes avec une spécificité concernant la couleur. Une image est le résultat d’interactions entre lumière et matière, nous tentons donc d’approfondir ces modèles physiques tout en maîtrisant les acquisitions lorsque cela est possible. Ensuite, une image ou une vidéo est un ensemble de données vectorielles, nous souhaitons le modéliser de façon optimale par des modèles mathématiques appropriés. Enfin, lorsque l’on traite une image, la finalité est souvent de la visualiser devant un observateur humain, d’où l’intérêt de le prendre en compte ainsi que son contexte. Nos spécificités reposent ainsi sur la définition d’outils mathématiques adaptés à la manipulation de données vectorielles, associée à l’introduction de paramètres physiques issus, d’une part, de modèles d’aspects des surfaces analysées, d’autre part, de modèles, du système visuel humain. De plus, notre originalité s’appuie sur la mise en place d’évaluations objectives et subjectives de la qualité, à la fois des traitements développés, mais aussi des supports de reproduction des images ou vidéos couleur, selon des conditions normalisées au sein de salles dédiées. La force et l’originalité de notre équipe repose donc sur le fait que nous travaillons sur toute la «chaîne de vie» d’une image couleur, depuis son acquisition jusqu’à l’évaluation de la qualité de son rendu, en passant par divers traitements; traitements définis de manière vectorielle et/ou sensorielle, en prenant éventuellement en compte les aspects physiques de la scène acquise.

Résumé de la thèse en français
À ce jour, il n’existe pas de méthode établie (ni subjective ni objective) pour garantir la qualité
de l’expérience pour le contenu 3D stéréoscopique quelle qu’en soit l’application (cinéma, télévision ou mobile). Ce travail de thèse est orienté vers l’amélioration et l’évaluation de la qualité des
contenus 3Ds. L’objectif est de produire des algorithmes et des modèles aidant à garantir
l’expérience de l’utilisateur étroitement liée à la perception.

Résumé de la thèse en anglais
o date there is no established methodology (neither subjective nor objective) to guarantee the quality of experience for stereoscopic 3D content whatever is the application (Cinema, television or mobile) . This PhD work is oriented towards quality enhancement and assessment of 3D content. The aim is to produce algorithms and models helping to guarantee the user experience closely linked to perception.

Description complète du sujet de thèse
The digital era allowed to simplify the spread of Stereoscopic 3D (S-3D) technologies in different application domains. The most important aspect is that S-3D can offer viewing experience providing sensations thanks to depth (binocular cues). Despites the challenge brought by S-3D technology, the latter has not yet been widely adopted. One can notice a decrease or speed-down especially for 3DTV. This mainly due to 3D-related issues that have been indentified on the whole chain. At the capture side, there are no common rules for creating a stereoscopic content besides some initiatives coming from stereographers not always leading to unanimity. Therefore, artifacts/defects may cause symptoms to viewers like headaches, nausea and visual fatigue. At the display side, if the projected content is not adapted to aspect ratio and, display size or technology, this will result in a decrease of the quality of experience.

Modern devices and networks are constructed for being used for S-3D content targeting applications such as video streaming, video-conferencing, camcording, etc. Moreover, high-dynamic-range Imaging together with S-3D is seen as an exciting feature towards augmented experience. However, to date there are no established methodology (neither subjective nor objective) to guarantee the quality of experience with such a content.

This PhD work is oriented towards quality enhancement and assessment of S-3D content on various devices (from cinema projector to mobile). The aim is to produce algorithms and models helping to guarantee the user experience. Five mains steps can be highlighted in this work:

1- Identification of factors affecting S-3D quality on various devices: S-3D has been addressed mostly for cinema but less for television and very rarely for mobile. Observation conditions are very different. Hence, it is important to understand intrinsic and extrinsic factors playing a role on quality. This part has de deal with the whole chain from content creation to display.

2- Perceptual model reflecting the effect of structural migrations on S-3D perception: It has been demonstrated in a previous work (Nauge 2012) that impairments can be interpreted as migration of statistical characteristics of an image. In this part, an extension will be proposed for S-3D video by taking into account spatial, temporal and inter-view migration. It will be followed by psychophysical experiments aiming at identifying links between stereo-spatio-temporal migration and perception. The obtained model will introduce a new framework for dealing with S-3D content and will the help in the definition of quality metrics and for compression enhancement.

3- Development of a quality procedure dedicated to S-3D and based on statistical modeling of impairments: The model resulting from phase 2 will be used for the construction of a 3D video metric that can be expressed depending of the availability of the reference. Of course, databases coming from extensive subjective evaluations will be created in order to validate the metrics and compare with the state of the art.

4- the next step consists in exploiting the aforementioned results for perceptual optimization of 3D extensions of HEVC known as the next standard for video coding. One possible aspect can be the use of statistical migrations in the rate-distortion optimization process.

5- Finally, possible extensions of the work can be to address higher dynamic range applied on 3D content or to consider hardware implementation of the developed algorithms in order to provide embedded systems based on FPGA for example.

Objectifs scientifiques de la thèse
This PhD work is oriented towards quality enhancement and assessment of S-3D content on various devices (from cinema projector to mobile). The aim is to produce algorithms and models helping to guarantee the user experience. Five mains steps can be highlighted in this work:

1- Identification of factors affecting S-3D quality on various devices: S-3D has been addressed mostly for cinema but less for television and very rarely for mobile. Observation conditions are very different. Hence, it is important to understand intrinsic and extrinsic factors playing a role on quality. This part has de deal with the whole chain from content creation to display.

2- Perceptual model reflecting the effect of structural migrations on S-3D perception: It has been demonstrated in a previous work (Nauge 2012) that impairments can be interpreted as migration of statistical characteristics of an image. In this part, an extension will be proposed for S-3D video by taking into account spatial, temporal and inter-view migration. It will be followed by psychophysical experiments aiming at identifying links between stereo-spatio-temporal migration and perception. The obtained model will introduce a new framework for dealing with S-3D content and will the help in the definition of quality metrics and for compression enhancement.

3- Development of a quality procedure dedicated to S-3D and based on statistical modeling of impairments: The model resulting from phase 2 will be used for the construction of a 3D video metric that can be expressed depending of the availability of the reference. Of course, databases coming from extensive subjective evaluations will be created in order to validate the metrics and compare with the state of the art.

4- the next step consists in exploiting the aforementioned results for perceptual optimization of 3D extensions of HEVC known as the next standard for video coding. One possible aspect can be the use of statistical migrations in the rate-distortion optimization process.

5- Finally, possible extensions of the work can be to address higher dynamic range applied on 3D content or to consider hardware implementation of the developed algorithms in order to provide embedded systems based on FPGA for example.

Compétences à l’issue de la thèse
3D quality assessment
Image description using natural scene statistics (NSS)
Enhancement of 3D coding
Hardware implementation of imaging algorithms

Mots clés (séparés par des virgules)
quality, S-3D, QoE, natural scene statistics, embedded systems
Conditions restrictive de candidature (nationalité, âge, …) : NON

Directeur de thèse
Chaker Larabi
Adresse mail du directeur de thèse : chaker.larabi@univ-poitiers.fr
Téléphone Directeur de thèse : +33549497494

Co-directeur de thèse
Faouzi Alaya Cheikh
Adresse mail du co-directeur de thèse : faouzi.cheikh@hig.no
Cofinancement LABEX SigmaLIM demandé : NON

Recherche

Menu principal

Haut de page