In this paper, a practical solution for optimal coding parameters using different bandwidth requirements is presented. The obtained specification is based on the analysis of a large database of more than 10,000 sequences compressed with different parameters. The obtained parameters can be used both for adaptive streaming or storage optimization.

1. Advanced Mechanisms for Delivering
High-Quality Digital Content
Mikołaj Leszczuk and Lucjan Janowski

2. Presentation outline
Conclusions
Results
Methods
Background

3. Conclusions

4. Current H.264 video codec mechanism
SouRCe
video
sequence
(SRC)
Codec
• Bit-rate
• Quantization
Parameter
(QP)
Processed
Video
Sequence
(PVS)

5. What is the problem?
• Limited coding parameters considered:
• Currently mainly:
• Bit-rate
• QP
• 10+ more to be considered
• Same parameters for all video sequences, while more detailed
content analysis possible using indicators like:
• Temporal Activity (TA)
• Spatial Activity (SA)
• Blur
• …
• How to map indicators onto coding parameters?

6. Methods

7. Video Quality Experts Group
• And here comes VQEG …
• Vision: “To advance the field of video quality
assessment...”
• International experts: industry, academia,
governmental, standard-developing
• VQEG’s Joint Effort Group (JEG)
• More information: http://www.vqeg.org/

8. Proposed H.264 video codec mechanism
SRC Analysis
• SA
• TA
• Blur
• …
Codec
• Bit-rate
• QP
• 10+
parameters
• …
PVS

9. What is our solution?
Selecting SRC test-set
Selecting number of
codec parameters
(Hypothetical
Reference Circuits,
HRC)
Employing super-
computer cluster to get
encoded (PVS) videos
Using existing well-
established objective
quality metrics for
ground-truth
Tuning results with
subjective
crowdsourcing and in-
lab experiments
Analysing indicators Get mapping
Developing codec
considering quality
factors

10. Selecting SRC
• 10×1080p@25
SRC
• Selected from VQEG
resources
• Covering many
different features:
• Natural and synthetic
• Professionally shot and
user generated
• Video content 

11. Selecting HRC (1/2)
Basic compression Temporal & spatial changes Time prediction I, B, P frame size factors
Bit-rate 1, 2, 4, 8, 16 Mbit/s
QP 26, 32, 38, 46
GOP length 8, 16, 32, 64 32, 64
Number of B frames 0, 2, 3, 7 2
B-pyramid strict, none none
Frame rate 25 8, 12 25
Resolution 1920×1080 960×540, 480×270 1920×1080
Integer pixel motion
estimation method
default dia, esa, umh default
Maximum motion vector
search range
default 4, 64 default
Number of reference frames default 4,16 default
Number of slices per frame 1, 2 1
I to P frame ratio default 0.8, 1, 1.2, 1.4
P to B frame ratio default 0.5, 0.8, 1, 1.2, 1.4

12. Selecting HRC (2/2)
Approximately
1300 distinctive
codec parameters!

13. Getting PVS (1/2)
• Pre-processing:
• Source AVI format using YCbCr space with 422
sampling
• Sub-sampled with Lanczos to 420 sampling
• Two well-known encoders: JM and x264
• Post-processing: decompression,
destination AVI format

14. Getting PVS (2/2)
Approximately
13,000 distinctive
PVS!

15. Analysis PVS with quality metrics
• Quality metrics used and inter-checked:
• Peak Signal-to-Noise Ratio (PSNR)
• Structural Similarity Index (SSIM)
• Video Quality Metric (VQM)
• Visual Information Fidelity (VIF)
• VQM shown to be best FR metric
• Fit factor of subjective data higher than other
metrics by ≥20%
• Therefore VQM metric used for further analysis of
obtained results

16. Analysing indicators
• Video indicators analysed: TA, SA, and
Blur
• More details on experiment reported in:
Leszczuk, M. et al., “Freely available
large-scale video quality assessment
database in full-HD resolution with
H.264 coding,” IEEE GLOBECOM
2013

17. Results

18. Getting mapping
Given video
sequence
Required
compression
rate
Best codec
parameters

19. Proof-of-Concept
• Developed advanced codec mechanism
• Supporting streaming on 5 popular browsers :
• Google Chrome (native support for HTML5/H.264)
• Mozilla Firefox (temporary solution)
• Microsoft Internet Explorer (temporary solution)
• Apple Safari (temporary solution)
• Opera (temporary solution)
• Plug-ins as temporary solutions:
• Windows Media Player (WMP) plug-in for Windows
• VideoLAN Client (VLC) for Linux and OS X

20. Integrated technologies
FFmpeg
Decoding
video input
(169 codecs)
x264
Parameterised
compression
itself
Matroska
Multimedia
container

21. Tested
solution
Multimedia
container
file inputs
Encoding
formats as
file input
Quality
models
Erroneous
input data

22. Conclusions

23. Resources available at
http://vq.kt.agh.edu.pl
Video
databases
Quality
metrics
Research
papers
Other
stuff