MPEG-21 Scalable Video MPEG-21 Scalable Video Coding Coding and the Stool (UniBS) system and the Stool (UniBS) system Riccardo Leonardi, Alberto Signoroni, Michele Brescianini, Nicola Adami Signals and Communications Lab. – DEA University of Brescia, Italy riccardo . leonardi @ ing . unibs .it alberto . signoroni @ ing . unibs .it Thessaloniki – October 7-8, 2004
18
Embed
MPEG-21 Scalable Video Coding and the Stool (UniBS) system Riccardo Leonardi, Alberto Signoroni, Michele Brescianini, Nicola Adami Signals and Communications.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
MPEG-21 Scalable Video MPEG-21 Scalable Video Coding Coding
and the Stool (UniBS) systemand the Stool (UniBS) system
Riccardo Leonardi, Alberto Signoroni,Michele Brescianini, Nicola Adami
Signals and Communications Lab. – DEAUniversity of Brescia, Italy
R. Leonardi - A. Signoroni University of Brescia, Italy
Scenarios (why SVC is desirable)MPEG Requirements (the “shall” and “should” of SVC)Technologies which enable SVCSVC architectures (t+2D, 2D+t, UniBS)Experimental results
OutlineOutline
R. Leonardi - A. Signoroni University of Brescia, Italy
Classical video coding Code-decode at a single operating point (target frame
resolution, spatial resolution and quality) Code-transcode at operating points different from the
target (complexity, performance loss, cost) Scalability in current standards (few spatio-temporal
operating points MPEG-2, coarse or fine grain quality scalability MPEG-4, performance loss)
Scalable video coding Code once decode at S x T x Q operating
points (full scalability in space, frame-rate and quality)
Video coding paradigmsVideo coding paradigms
R. Leonardi - A. Signoroni University of Brescia, Italy
Multi-channel content production and delivery: the same stream will be viewed on a variety of devices
with different spatial and/or temporal resolution will be viewed and charged on the basis of delivered
quality level devices can decide to trade-off received quality for a
longer battery life (complexity scalability)
Video streaming over heterogeneous IP networksAdvanced surveillance systemsVideo conferencingMobile streaming videoOthers...
SVC scenarios and SVC scenarios and applicationsapplications
R. Leonardi - A. Signoroni University of Brescia, Italy
at Pattaya Meeting (Dec.2001) MPEG started an AHG to explore Interframe Wavelet Coding – activity successfully completed in Oct.2002RPI software (prof. J.W.Woods) released based on MCTF+EZBC
MPEG started an AHG on Scalable Video Coding (MPEG-21)Call for Evidence on SVC: 9 responses received by July 2003Call for Proposals on SVC: 14 full proposals (Wavelet based, AVC based)
Wavelet based: best selected Microsoft Research Asia (software released)
AVC based: best selected HHI-FraunhoferCore Experiments started at the Monaco (Mar.2004) and Redmond (Jul.2004) meetings.No reference architecture yet specified (MRSA, HHI, Samsung, UNIBS, DANAE-TOMPHSON, UNSW proposals at the next meeting Palma de Mallorca Oct.04)
MPEG-21 Ad Hoc Group on MPEG-21 Ad Hoc Group on SVCSVC
R. Leonardi - A. Signoroni University of Brescia, Italy
RequirementsRequirements
shall should may
Spatial scalability
2 levels(e.g.
CIF+QCIF)4 levels
From QQCIF to
3610x1536
Temporal scalability
2 levels4 levels
(up to 60 Hz)
Non-dyadic
Quality (SNR)
scalability
Coarse grain (25% bit-
rate)
Medium grain
(10% bit-rate)
FGS or progressive
Last revision: MPEG output doc N6505, Redmond, July 2004
R. Leonardi - A. Signoroni University of Brescia, Italy
Combined Scalability (at least for a finite number of points in the spatial-temporal-quality space)Challenging Reqs
Coding Efficiency performance (superior to H.264/AVC)
Base-layer compatibility (to H.264/AVC)
Complexity scalability (for power limited devices)
Others (less peculiar to SVC) ROI and object scalability, robustness to transmission
errors and on “best-efforts” networks, end-to-end delay, random access capability)
Requirements Requirements contd.
R. Leonardi - A. Signoroni University of Brescia, Italy
Scenario 1 – merged proposal for sub-extractionsCity, Crew, Harbour, Soccer
R. Leonardi - A. Signoroni University of Brescia, Italy
A generic SVC encoderA generic SVC encoder
Video
Entropy Coding
Motion Coding
Texture Coding
Spatio-Temporal"Transform"
2D SpatialTransform
TemporalTransform
Bitstream
In order to have an efficient scalable codec, it is necessary that In order to have an efficient scalable codec, it is necessary that each one of the building blocks is operating in a scalable mode.each one of the building blocks is operating in a scalable mode.
MPEG-21 doc N6520 Scalable Video Model 2.0 July 2004, Redmond, WA, USA
R. Leonardi - A. Signoroni University of Brescia, Italy
In the MPEG21-SVC framework a series of SVC enabling technologies have been identified
R. Leonardi - A. Signoroni University of Brescia, Italy
SVC implementationsSVC implementations
• Following the Call for Proposal for SVC (MPEG doc.N5958 Oct.2003) a certain number of implementations have been presented (MPEG doc.M10569 Mar.2004).
• On the base of the CfP inputs a SVC classification has been made which allows to distinguish some families of SVC implementations:
“t+2D” schemes “2D+t” schemes “2D+t+2D” schemes or “Multi-scale Pyramid t+2D”
Our STool scheme could be classified in the third category
R. Leonardi - A. Signoroni University of Brescia, Italy
t+2D schemest+2D schemes
Video Bitstream
MultiplexMCTF
Motion Coding(Scalable)
Motion
TextureSNR Scalable
Entropy CodingDWT (N level) HP/LP
Enablestemporal scalability
Enablesspatial scalability
Enablesquality scalability
R. Leonardi - A. Signoroni University of Brescia, Italy
2D+t schemes2D+t schemes
Video Bitstream
DWT (1 level) MultiplexMCTF
Motion CodingMotion
TextureSNR Scalable
Entropy CodingHP
DWT (1 level)
LP
LP
MCTF
Motion CodingMotion
TextureSNR Scalable
Entropy CodingHP
MCTF
Motion CodingMotion
Texture SNR ScalableEntropy Coding
DWT(N-3 level)
HP/LP
R. Leonardi - A. Signoroni University of Brescia, Italy
R. Leonardi - A. Signoroni University of Brescia, Italy
Results: Results: CITY QCIF 15 Hz 64 kbps
Original AVC STool
zoom zoom
R. Leonardi - A. Signoroni University of Brescia, Italy
Results:Results:
Original
AVC Stool
CITY CIF 30 Hz 256 kbps
R. Leonardi - A. Signoroni University of Brescia, Italy
Results: AVC vs MSRA vs Results: AVC vs MSRA vs StoolStool
Play
HARBOUR
CITY
QCIF 15Hz 96kbps
CIF 30Hz 384kbps
QCIF 15Hz 64kbps
CIF 30Hz 256kbps
R. Leonardi - A. Signoroni University of Brescia, Italy
ConclusionsConclusions
SVC systemsMultipoint spatial-temporal-SNR scalabilityComparable or superior in coding performance with respect to AVCA lot of work is currently made on SVC (AVC is mature, SVC is young… :-)
SToolSTool: a new 2D+t+2D system with good performance (UNIBS submission to the next MPEG meeting)