Multiview Coding and Compression for 3D Video

Résumé

Multiview video (MVV) is an advanced representation of 3D video technology. The MVV content depends on several parameters such as the deployed cameras number, recording angles and the captured scene. Each used camera inherently generates an extra amount of data compared to the 2D conventional video that only needs one camera. MVV needs specific coding techniques that take into consideration the visual similarities between the viewpoints set. Both temporal and interview correlations could be exploited to improve the compression ratios. Multiview video coding (MVC) is the extended profile of H.264/AVC video codec. MVC offers compression efficiency improvements achieving 50 % over simulcast video coding. However, better compression performance leads to higher random access complexity which might hamper MVC usage in applications such as Free viewpoint Television and 3D TV broadcasting. It also degrades the viewers quality of experience due to the lack of interactivity. This thesis aims to study MVC encoding and to solve the aforementioned problem of Random Access (RA) ability by proposing faster interview prediction approaches. The thesis first presents a brief description of the 3D video concepts from capturing to displaying. Furthermore, it focuses on video coding fundamentals and its multiview extension form. The thesis then proposes two novel techniques to enhance the random access ability of the MVC encoder. The first proposed approach (PBI) aims to lower the cost of randomly accessing any picture at any position and instant, with respect to the multiview reference model JMVM and other state-of-the-art methods. The proposed (PBI) scheme is mainly based on the use of two base views (I-views) in the interview structure with selected positions instead of only a single reference view as in the standard structure. This provides a direct interview prediction for the remaining views and ensures faster random access ability while maintaining a competitive compression performance. PBI achieves a random access gain of 20 % relative to the reference model MVC. The second proposed approach (PIP) surpasses PBI structure by achieving a random access gain of 53.33% compared to the benchmark standard MVC. A novel random access ability evaluation method (GR) has been suggested and adopted throughout the thesis experimental parts. It allows more accurate assessment by considering all pictures types of the tested multiview schemes. A comparative investigation of proposed and reported multiview schemes is also presented in this thesis. Results of the conducted tests allow classifying the examined multiview schemes and GOP sizes preferences according to their effects in terms of random access ability and compression efficiency. Finally, more experiments have been conducted comparing MVC and MV-HEVC standards in terms of compression efficiency using different multiview video sequences that have different textures and resolutions.

Citer ce document

Accès au document

Ce document est hébergé sur son dépôt institutionnel d'origine.

Auteur(s)

Statistiques

Consultations : 2

Téléchargements : 0