1. CS644 Advanced Topics in Networking
Multimedia
Mon, 2004/9/13

2. Video Basics Compression Performance issues Available tools
Lossless compression
MPEG-1,2,4, and 7
Performance issues
Old & new
Available tools

3. Video Basics: Representations
RGB
each pixel per red, green, blue
YUV
Y: luminance
UV: chrominance
better match the human visual system
can distinguish the luminance (brightness) better than its hue (color)

4. Compression Techniques
Lossless
MPEG-1,2,4, adn 7

5. Lossless Compression Run Length Encoding
AAABBCDDD => 3A2B1C4D
Difference Pulse Code Modulation
AAABBCDDD => A
Dictionary-based
Lempel-Ziv (LZ)
GIF: uses standardized colormap
Huffman Coding

6. Huffman Coding
0 01
1 001
3 000
4 1 1 1 4 4 3 3 4 4 4 4
0 1
To perform Huffman coding, the first step is to use the frequency statistics to generate a prefix code representation of the indices, such as the one shown here. Notice how the more frequently appearing symbols are represented with a shorter code.
All that is left then is to output the bit stream corresponding to the codeword representations of the indices, as illustrated here.
As you can see, the compression pipeline is considerably more complicated than a simpler encoder, but the effort we spent in modeling the data means that we can reduce the input redundancies to the best of our capabilities.
We can perform a simple analysis for the efficiency of this scheme. The input string consists of 6 characters, each of 8 bits in length, so 48 bits in total. The final representation of this information is only 13 bits, compressed to almost 4 times the original size. As you can see, even with this trivial example, the amount of reduction is no less than remarkable.
To summarize, the Burrows-Wheeler compression procedure consists of four stages of transformation. The first stage aims to reduce simple repetitions of input symbols, and the second stage transforms the data by grouping symbols with similar context. The third stage utilizes the locality of symbols in the sorted block, and finally, the entropy coding stage outputs optimal representation of the original data. The reverse transform unravels the entropy-coded information in a similar manner and will restore the original data without any loss.
Input = 6 x 8 = 48 bits
Output = 13 bits
Compression = 73 %

7. Wavelets
“Mathematical functions that cut up data into different frequency components and study each component with a resolution matched to its scale”

8. Haar transform
decomposes a discrete signal into two subsignals of half its length
f = (4,6,10,12,8,6,5,5)
(a|d) = (5x21/2, 11x21/2 7x21/2, 5x21/2| -21/2, -21/2, 21/2, 0)
E = f12+ f22+ f f82 = 446
E(a) = 440, E(d) = 6

9. Haar transform

11. MPEG
Moving Picture Experts Group in ISO/IEC specifies:

12. Multimedia Technology Standards
Basic Set and
Animation/graphics/text
Audio/visual objects composed in 3-D space
Allows generic management of delivery systems
Basic Set
Source
Delivery
Demultiplexer
Audio/video
MPEG - 4
MPEG - 7
Current challenge:
Specify standard set of descriptors for various types of MM information
MPEG - 1
MPEG - 2
Basic Set and
supporting high quality and interlace video

13. MPEG-1
Started on 1988
A/V

14. MPEG-2 A/V for broadcast 4~15 Mbps for digital TV
digital TV, HDTV, cable, satellite
4~15 Mbps for digital TV

15. MPEG-4 Goals for natural and synthetic video object-oriented coding
low-bit rate for wireless
MPEG-2 backward compatibility
DCT -> Quantization -> Encoding

16. Compression:Temporal Redundancy
Without B-frame
With B-frame

17. MPEG-7 Goals Officially, “Multimedia Content Description Interface”
Allow efficient search for multimedia content that is of interest to user using standardized descriptions
MPEG-1,2,4: representation of content itself (the bit)
MPEG-7: representation of information about the content (the bit about the bits)

18. What is Exactly Included?
Not the analysis
Not the search engine
Just the description
Feature Extraction
Description
Search Engine
This is the scope
of MPEG-7

19. MPEG-7 Components Descriptors (Ds) Description Schemes (DSs)
representations of features
define the syntax and semantics
Description Schemes (DSs)
specify structure and semantics of the relationships between components (Ds and DSs)
Description Definition Language (D이)
allow creation of new DSs and Ds
allow extension and modification of DSs

20. Example of a client-server architecture in a MPEG-7 based data search
Initial Descriptor Table
text link
feature selection
Interface
Interface
Search
Engine
MPEG-7 stream
MPEG-7
Decoder
MPEG-7 database
Query result
(link address)
AV content
Interface
Interface
Presentation
Engine
AV content
AV data
Decoder
AV stream
AV content

21. Networking Issues Different from VoIP
Higher b/w requirements
No silence periods
Complex codecs
More loss tolerant

22. Video Streaming Old Issues Current Issues Variable Bit Rate
Server-side transmission scheduling
Network provisioning
Current Issues
Multimedia in “wireless” environment
Batter-power limited, diverse technologies, low-layer dependent, application-dependent, etc.

23. Acknowledgements
Kave Salamtian’s “Multimedia Storage and Compression”