Optimisation (perceptuelle) d'HEVC et des futurs standards de compression vidéo.

Nous vous rappelons que, afin de garantir l'accès de tous les inscrits aux salles de réunion, l'inscription aux réunions est gratuite mais obligatoire.

Inscriptions closes à cette réunion.

Inscriptions

8 personnes membres du GdR ISIS, et 17 personnes non membres du GdR, sont inscrits à cette réunion.
Capacité de la salle : 50 personnes.

Annonce

Annonce :

• Titre : « Optimisation (perceptuelle) d'HEVC et des futurs standards de compression vidéo ».
• Date : 8 juillet 2016
• Lieu : à Polytech'Nantes, dans le bâtiment IHT, salle Nemo (A102)
• Adresse : La Chantrerie, rue Christian Pauc, Nantes.

Organisateurs :

• Vincent Ricordel (IRCCyN Nantes, UMR 6597)
• Wassim Hamidouche (IETR Rennes)

Description et appel à communications :

La recherche en compression vidéo est toujours conduite avec l'objectif de délivrer des volumes élevés de contenus plus immersifs sur des canaux à bande limitée voire congestionnés.

L'objectif est aussi que les contenus décodés possèdent une qualité subjective irréprochable (on parle aussi de fidélité perceptuelle sans perte).

Cette journée du GDR sera donc autour du codage vidéo et s'intéressera de façon large aux approches d'optimisation d'HEVC et des futurs standards de compression vidéo.

Nous appelons donc des présentations relatives :

• aux approches d'optimisations (débit-qualité, transformations, métriques, évaluation de la qualité, ...)
• aux outils et méthodes d'adaptations des codeurs (synthèse de textures, couleur, scalabilité, multi-vues, 3D, dynamiques étendues, ...)
• à l'adéquation au hardware.

Les personnes souhaitant présenter leurs travaux sont priées d'envoyer un résumé d'une page maximum (format pdf) avant le 1er mai 2016 aux organisateurs :

• Vincent.Ricordel@univ-nantes.fr
• et Wassim.Hamidouche@insa-rennes.fr

Intervenant invité :

Enrico Magli (IEEE distinguished lecturer)
http://staff.polito.it/enrico.magli
Titre : « Advances in Image/Video Compression Using Graph Transforms ».

Cette journée est organisée avec le soutien :

• du GDR ISIS
• de l'IEEE Circuits and Systems Society (Distinguished Lecturer Program)
• du projet européen PROVISION ITN<

Programme

9:15 - 9h45 : Accueil

9:45 - 10h00 : Présentation introductive (GdR ISIS)

10h00 - 11h00 : Présentation 1 - invité
- Enrico Magli (IEEE DLP, Politecnico di Torino)
- Titre : Advances in Image/Video Compression Using Graph Transforms

11h00 - 11h30 : Présentation 2
- Sébastien Lasserre (Technicolor, Rennes)
- Titre : SDR color optimization relatively to original HDR produced contents.

11h30 - 11h45 : Pause café

11h45 - 12h15 : Présentation 3
- Karam Naser (IRCCyN, Nantes)
- Titre : Estimation of perceptual redundancies in HEVC encoded dynamic textures.

12h15 - 12h45 : Présentation 4
- Antoine Dricot (Telecom ParisTech)
- Titre : Subjective Evaluation of Super Multi-View Compressed Contents on High-End Light-Field Displays.

12h45 - 14h15 : Déjeuner

14h15 - 14h45 : Présentation 5
- Glenn Herrou (b-Com, Rennes)
- Titre : Performance Comparison of HEVC and VP9 for HDR Video Coding.

14h45 - 15h15 : Présentation 6
- Thibaud Biatek (INSA Rennes - b-Com - TDF)
- Titre : Optimal Bitrate Allocation in the Scalable HEVC Extension for the Deployment of UHD Services.

15h15 - 15h30 : Pause café

15h30 - 16h00 : Présentation 7
- Olivier Le Meur (IRISA, Rennes)
- Titre : Saccadic model, a promising framework for modelling overt visual attention.

16h00 - 16h30 : Présentation 8
- Yashas Rai (IRCCyN, Nantes)
- Titre : Naturalistic measurement of perceived video quality by measuring disruptions in free-viewing gaze patterns.

17h00 - 17h30 : Présentation 9
- Victorien Lorcy (b-Com, Cesson-Sévigné)
- Titre : Transform coding for future video codecs.

17h30 - 18h00 : Discussion et Conclusion

Résumés des contributions

10h00 - 11h00 : Présentation 1 - invité

• Enrico Magli (IEEE DLP, Politecnico di Torino)
• Titre : Advances in image/video compression using graph transforms
• Résumé :
  This talk describes recent developments in several areas of image and video compression employing new signal processing and computer vision techniques. After an initial discussion about the future of video compression, the talk will be centered around graph transforms. I will first address the design of graph transforms achieving very good compression efficiency on both piecewise-smooth and natural images, and their application to image compression and intra-frame video compression. Next, I will introduce a new video compression paradigm that replaces macro blocks with superpixels, and employs a graph transform to encode each super pixel. Finally, I will describe a new and very efficient transform, called ?steerable? DCT, which is able to orient the DCT basis functions spatially and spectrally so as to adapt to directional features of the block being encoded.

11h00 - 11h30 : Présentation 2

• Sébastien Lasserre (Technicolor, Rennes)
• Titre : SDR color optimization relatively to original HDR produced contents.
• Résumé :
  Ultra HD TV (UHDTV) is going to enter the broadcasting market soon as a major quality improvement compared to HDTV. Beyond more pixels (from 2k to 4k definition) and faster pixels (higher frame rate), the main improvement is expected to be provided by a higher dynamic range (HDR) and more saturated colors using a Wide Color Gamut (WCG). Basically, HDR and WCG are two different ways at considering a bigger color volume. For the broadcast market, it is highly desirable to be able to distribute both Standard Dynamic Range (SDR) and HDR versions of a content without having to multicast the distribution as it would virtually double the needed bandwidth. Technicolor has developed a unique solution that allows the distribution of an HDR content and a SDR version of this content using a single HEVC stream. The presentation will focus on color optimization of the generated SDR version relatively to the original HDR produced content. It will be demonstrated how a complete theoretical color volume analysis can lead to a very simple and practical tone mapper operator, from HDR to SDR, that is mathematically invertible.

11h45 - 12h15 : Présentation 3

• Karam Naser (IRCCyN, Nantes)
• Titre : Estimation of perceptual redundancies in HEVC encoded dynamic textures.
• Résumé :
  Statistical redundancies have been the dominant target in the image/video compression standards. Perceptually, there exists further redundancies that can be removed to further enhance the compression efficiency. In this presentation, we considered short term homogeneous patches that fall into the foveal vision as dynamic textures, for which a psychophysical test was used to estimate their amount of perceptual redundancies. We demonstrated the possible rate saving by utilizing these redundancies. We further designed a learning model that can precisely predict the amount of redundancies and accordingly proposed a generalized perceptual optimization framework.

12h15 - 12h45 : Présentation 4

• Antoine Dricot (Telecom ParisTech)
• Titre : Subjective Evaluation of Super Multi-View Compressed Contents on High-End Light-Field Displays.
• Résumé :
  Super Multi-View (SMV) video content is composed of tens or hundreds of views that provide a light-field representation of a scene. It allows a glass-free visualization and eliminates many causes of discomfort existing in current available 3D video technologies. Efficient SMV video compression is a key factor for enabling future light-field video services. In this presentation, we first compare several coding configurations, both objectively and subjectively. We assess the impact of view synthesis on the coding scheme, and report ranges of bitrates required to obtain a good quality for the tested content. The reliability of the PSNR metric for SMV content is also studied. Finally, preliminary observations are initiated about the light-field conversion step and about the perception of motion parallax. To the best of our knowledge, this work is the first to carry out subjective experiments and to report results of SMV compression for light-field displays. It provides first results showing that improvement of compression efficiency is required, as well as depth estimation and view synthesis algorithms improvement, but that the use of SMV appears realistic according to next generation compression technology requirements.

14h15 - 14h45 : Présentation 5

• Glenn Herrou (b-Com, Rennes)
• Titre : Performance Comparison of HEVC and VP9 for HDR Video Coding
• Résumé :
  Current increasing effort in the television industry towards High Dynamic Range (HDR) imaging has raised the issue of the compression of HDR content. Offering a higher peak luminance and wider color gamut, HDR content introduces new challenges to the state-of-the-art video codecs such as High Efficiency Video Coding (HEVC) or VP9, which have been designed for Standard Dynamic Range (SDR) video coding. This study presents a performance comparison between HEVC and VP9 in the HDR context through both objective and subjective evaluations. The experimental results have shown that HEVC offers from 0.6% to 38.2% bit rate savings over VP9 depending on the used objective metric. The subjective study showed that bit rate savings greater than 38.7% on average can be achieved by HEVC for the same perceived quality as VP9.

14h45 - 15h15 : Présentation 6

• Thibaud Biatek (INSA Rennes - b-Com - TDF)
• Titre : Optimal Bitrate Allocation in the Scalable HEVC Extension for the Deployment of UHD Services
• Résumé :
  Ultra High Definition (UHD) is the latest trend in broadcasting area, which enables new services with 3840x2160 resolution and comes with enhanced color-gamut, frame-rate, dynamic range and better audio system compared to the currently deployed HD services. The UHD format for broadcasting is already under standardization in the DVB consortium which plans to introduce UHD services in three phases. The increase in data brought by these services requires more efficient compression and transmission systems. The recent scalable video coding standard SHVC is a promising candidate to handle these three phases while ensuring backward compatibility. Moreover, delivering such contents over networks needs an accurate control of the output bitrate from encoder engines to match rigid constraints on bandwidth and QoS. Several contributions have already been proposed to jointly encode scalable stream, but without considering the impact of bitrate ratio between layers on the compression performance. In this paper, we first investigate the impact of the bitrate ratio between layers on the coding performance for several UHD scalable schemes including spatial, color-gamut and SDR-to-HDR scalability in SHVC. Based on this investigation, we propose an adaptive rate control algorithm which dynamically allocates the bitrate between two layers to optimize the performance under quality and bitrate constraints. The algorithm has been implemented in the latest SHVC reference software (SHM9.0) and tested on 15 video sequences within two industrial use-cases. The performance shows an average BD-BR improvement of 7.51% and 3.35% for these use-cases.

15h30 - 16h00 : Présentation 7

• Olivier Le Meur (IRISA, Rennes)
• Titre : Saccadic model, a promising framework for modelling overt visual attention.
• Résumé :
  When looking at complex visual scenes, we perform in average 4 visual xations per second. This dynamic exploration allows selecting the most relevant parts of the visual scene and bringing the high-resolution part of the retina, the fovea, onto them. Saccadic models strive to overcome the limitations of existing saliency models. Rather than computing a unique saliency map, saccadic models aim to predict visual scanpaths, i.e. the suite of xations and saccades an observer would perform to sample the visual environment. As saliency models, saccadic models have to predict the salient areas of our visual environment. But the great difference with saliency models is that saccadic models have to output plausible visual scanpaths, i.e. having the same peculiarities as human scanpaths. I will present the saccadic model we recently published in Vision Research and underline why this new framework is promising for predicting eye fixations.

16h00 - 16h30 : Présentation 8

• Yashas Rai (IRCCyN, Nantes)
• Titre : Naturalistic measurement of perceived video quality by measuring disruptions in free-viewing gaze patterns.
• Résumé :
  Eye movements are principally driven by two factors: bottom up content features and top-down perceived interest that is based on our semantic interpretation of the scene. Eye-movements and Visual attention are therefore strongly influenced by the Supra-Threshold distortions present in the video. Quantifying these disruptions provides us with a naturalistic measure of scene quality without any unnatural examination of the scene or attention modulation. By training a Gaussian HMM model, using the data obtained from a carefully controlled subjective experiment, we are able to quantify the extent to which disruptions alter natural viewing patterns. In the current analysis, we gracefully degrade the quality away from the viewpoint of the observer, in order to achieve optimum viewing experience with potentially large bitrate savings. We confirm that these HEVC coding distortions in the visual periphery have no significant impact in case of both: para and per-foveal cases.

17h00 - 17h30 : Présentation 9

• Victorien Lorcy (b-Com, Cesson-Sévigné)
• Titre : Transform coding for future video codecs.
• Résumé :
  The transform operation like Discrete Cosinus Transform (DCT) is a core step in state-of-the art video codecs. It enables to decorrelate the signal and compacts its energy on a low number of transform coefficients. The latest video standard, High Efficiency Video Coding (HEVC), includes two different trigonometric transforms : the type II DCT and the type VII DST. Recently, MPEG and ITU jointly launched a group to investigate possible improvements of coding efficiency over HEVC under the Joint Video Exploration Team (JVET). A new approach of transform coding has been introduced in JVET namely the Adaptive Multiple Core Transforms (AMT). AMT adds 3 new transforms to the ones used in HEVC, forming a set of 5 trigonometric transforms : type II DCT, type VII DST, type V DCT, type VIII DCT, type I DST. AMT allows for selection of transform (in a RDO fashion) in a predefined set, which is then signalled in the bitstream. Bit rate reduction of -2.8 % are achieved over HEVC for the luma component with AMT. The presentation will concern AMT in its latest version and the future of transform coding for video compression.