Scalable Video Coding Dept Talk Ankit Bhurane

47
Scalable Video Coding: An Overview Ankit A. Bhurane TIDSP, EE, IITB

description

Scalable_Video_Coding_Dept_Talk_Ankit_Bhurane

Transcript of Scalable Video Coding Dept Talk Ankit Bhurane

Page 1: Scalable Video Coding Dept Talk Ankit Bhurane

Scalable Video Coding: An Overview

Ankit A. Bhurane

TIDSP, EE, IITB

Page 2: Scalable Video Coding Dept Talk Ankit Bhurane

Why Video?

• Annual global IP traffic will surpass 1.3 zettabytes/ month threshold by the end of 2016!

• 1, 000, 000, 000, 000, 000, 000, 000 bytes!

• Who is responsible?

• Complete list at http://en.wikipedia.org/wiki/List_of_video_hosting_services

Forecast by CISCO: http://www.cisco.com/

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 2

Page 3: Scalable Video Coding Dept Talk Ankit Bhurane

Raw Images

• Store data from an image sensor

• Typical image sensors capture only red, green, or blue for a pixel.

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 3

Courtesy: Adobe Systems

Page 5: Scalable Video Coding Dept Talk Ankit Bhurane

How Hard can it be?

• Size: HD: 1280×720;

• Color

• Frame rate: 30 fps

• 1280×720×3×30×8 bits/sec

• 663.552000 Mbits/sec!

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 5

Page 6: Scalable Video Coding Dept Talk Ankit Bhurane

Raw Video Specs

• Specification:

– Spatial Resolution (aspect ratio)– 4:3, 16:9, …

– Color space: RGB, YUV, …

– Framerate: 30 fps, 50 fps, 60 fps

– ?

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 6

Page 7: Scalable Video Coding Dept Talk Ankit Bhurane

Resoltion Chart: Source: Wikipedia

http://upload.wikimedia.org/wikipedia/en/9/9b/Vector_Video_Standards2.png

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 7

Page 8: Scalable Video Coding Dept Talk Ankit Bhurane

Frame rate

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 8

Page 9: Scalable Video Coding Dept Talk Ankit Bhurane

Color Space

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 9

Page 10: Scalable Video Coding Dept Talk Ankit Bhurane

RGB to YUV

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 10

RGB YUV

𝑌𝑈𝑉

=0.299 0.587 0.114−0.147 −0.289 0.4360.615 −0.515 −0.100

𝑅𝐺𝐵

Page 11: Scalable Video Coding Dept Talk Ankit Bhurane

How to achieve compression?

• Take advantage of human visual system (HVS)!

• Remove redundancy.

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 11

Page 12: Scalable Video Coding Dept Talk Ankit Bhurane

Human Eye Sensitivity

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 12

Shamelessly Taken from: Slides by Yao Wang, Polytechnic University, Brooklyn :P

Courtesy: Yao Wang

Page 13: Scalable Video Coding Dept Talk Ankit Bhurane

Croma Sampling

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 13

Shamelessly Taken from: Slides by Yao Wang, Polytechnic University, Brooklyn :P

Courtesy: Yao Wang

Page 14: Scalable Video Coding Dept Talk Ankit Bhurane

Most video is 4:2:0

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 14

Taken from: http://people.xiph.org/~tterribe/pubs/lca2008/anatomy.pdf

Page 15: Scalable Video Coding Dept Talk Ankit Bhurane

Spot the difference

Department Talk | Ankit Bhurane | TIDSP-IITB April 3, 2013 15

4:4:4 4:2:2

Sequences available at: http://samples.mplayerhq.hu/yuv/

4:2:2 requires 2/3rd the bandwidth of 4:4:4!

Page 16: Scalable Video Coding Dept Talk Ankit Bhurane

How to achieve compression?

• Take advantage of human visual system!

• Remove redundancy.

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 16

Page 17: Scalable Video Coding Dept Talk Ankit Bhurane

Spatial Redudndancy: Correlated data

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 17

Page 18: Scalable Video Coding Dept Talk Ankit Bhurane

Transform

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 18

Taken from: http://people.xiph.org/~tterribe/pubs/lca2008/anatomy.pdf http://www.vcodex.com/files/videocoding_intro_1.pdf

Page 19: Scalable Video Coding Dept Talk Ankit Bhurane

Spatial-Temporal redundancy

n n+1

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 19

Page 20: Scalable Video Coding Dept Talk Ankit Bhurane

MEMC

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 20

Taken from: John G. Apostolopoulos, Image and Video Compression, MIT 6.344, Spring 2002

Motion

Page 21: Scalable Video Coding Dept Talk Ankit Bhurane

Motion Estimation and

Compensation

Department Talk | Ankit Bhurane | TIDSP-IITB April 3, 2013 21

Taken from: Thyagarajan, Still Image and Video Compression.

Page 22: Scalable Video Coding Dept Talk Ankit Bhurane

Department Talk | Ankit Bhurane | TIDSP-IITB April 3, 2013 22

Page 23: Scalable Video Coding Dept Talk Ankit Bhurane

MEMC-Example

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 23

Page 24: Scalable Video Coding Dept Talk Ankit Bhurane

How to achieve compression?

• Take advantage of human visual system!

• Remove redundancy.

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 24

Page 25: Scalable Video Coding Dept Talk Ankit Bhurane

Campbell-Robson

Contrast Sensitivity Chart

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 25

Sinusoidally modulated Luminance

Co

ntr

ast

fro

m 1

00

% t

o a

bo

ut

0.5

%

http://neurovision.berkeley.edu/Demonstrations/VSOC/izumi/CSF/A_What_is_CSF.html

Page 26: Scalable Video Coding Dept Talk Ankit Bhurane

Quantization

• The only lossy step!

• Broadly categorized:

– Scalar

– Vector

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 26

Page 27: Scalable Video Coding Dept Talk Ankit Bhurane

Core of Data Compression

Transform

• Removes spatial redundancy

Quantize

• Removes insignificant data

Entropy coding

• Removes statistical redundancy

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 27

Page 28: Scalable Video Coding Dept Talk Ankit Bhurane

Core of Video Compression

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 28

MEMC

• Removes temporal redundancy

Transform

• Removes spatial redundancy

Quantize

• Removes irrelevant data

Entropy coding

• Removes statistical redundancy

Page 29: Scalable Video Coding Dept Talk Ankit Bhurane

Transform

• Goal: A sparse representation of the pixel data.

– By removing redundancy of highly correlated data.

• Optimal transform: Karhunen-Loève Transform (KLT).

– Data-dependent

– Need to transmit eigen vectors

• More appropriate: DCT

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 29

Page 30: Scalable Video Coding Dept Talk Ankit Bhurane

Transform

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 30

Taken from: http://people.xiph.org/~tterribe/pubs/lca2008/anatomy.pdf http://www.vcodex.com/files/videocoding_intro_1.pdf

Page 31: Scalable Video Coding Dept Talk Ankit Bhurane

Entropy Coding

• Inputs:

– transform coefficients

– motion vectors and

– ‘side’ information (headers, synchronization markers, etc.)

• Techniques:

– Huffman coding

– Arithmetic coding

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 31

Page 32: Scalable Video Coding Dept Talk Ankit Bhurane

Entropy Coding

• Huffman coding

– Optimal– but only as symbol codes

– Needs integer-valued codeword length

• Arithmetic coding

– Codes group/ block of symbols

– Near-optimal for long data

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 32

Page 33: Scalable Video Coding Dept Talk Ankit Bhurane
error.swf
Page 35: Scalable Video Coding Dept Talk Ankit Bhurane

How it comes to us?

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 35

Internet

www.youtube.com 208.65.153.251

Server

Internet Service Provider

Page 36: Scalable Video Coding Dept Talk Ankit Bhurane

How does it look in Market?

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 36

Page 37: Scalable Video Coding Dept Talk Ankit Bhurane

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 37

Todays Need

Page 38: Scalable Video Coding Dept Talk Ankit Bhurane

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 38

Encode: Once at highest resolution, Decode: Different rates depending on application

Single bit stream adaptable to heterogeneous end users!

Scalability, What’s that?

Page 39: Scalable Video Coding Dept Talk Ankit Bhurane

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 39

Scalable Bitstream

Courtesy: Nikola Adami et al., “State-of-the art and trends in Scalable video compression with wavelet-based approaches”

Page 40: Scalable Video Coding Dept Talk Ankit Bhurane

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 40

SVC System

Courtesy: Nikola Adami et al., “State-of-the art and trends in Scalable video compression with wavelet-based approaches”

Page 41: Scalable Video Coding Dept Talk Ankit Bhurane

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 41

Scalable Bitstream

Courtesy: Nikola Adami et al., “State-of-the art and trends in Scalable video compression with wavelet-based approaches”

Page 42: Scalable Video Coding Dept Talk Ankit Bhurane

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 42

Spatial Scalability

Page 43: Scalable Video Coding Dept Talk Ankit Bhurane

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 43

Temporal Scalability

Page 44: Scalable Video Coding Dept Talk Ankit Bhurane

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 44

Quality Scalability

Page 45: Scalable Video Coding Dept Talk Ankit Bhurane

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 45

Performance

Page 46: Scalable Video Coding Dept Talk Ankit Bhurane

April 3, 2013 Department Talk | Ankit Bhurane | TIDSP-IITB 46

SVC demo

Courtesy: Georgia Tech.

Page 47: Scalable Video Coding Dept Talk Ankit Bhurane