Video compression techniques used in broadcasting

1. Video Compression

2. Video Compression
 Outline
 The need for compression
 Types of redundancy
 Image compression
 Video compression

3. The need for compression
 Raw video contains an immense amount of
data
 Communication and storage capabilities are
limited and expensive
 Example HDTV video signal:
 1280X720 pixels/frame, progressive
scanning at 60 frames/s:
 20 Mb/s HDTV channel bandwidth
 Requires compression by a factor of 70:1

4. Types of redundancy
 Spectral Redundancy
 Spatial redundancy (Intra – frame compression)
 Temporal redundancy (Inter – frame compression)
 Entropy Redundancy (Lossless compression)
 Psycho-visual redundancy

5. Spectral Redundancy
 The RGB signals from video cameras are highly
correlated and take large bandwidth of 15 MHz
 To decrease the amount of video sample data
based on human perception, the RGB color space is
converted to Y, Cr & Cb color space
 The Y has the full bandwidth as it is very sensitive
to human perception
 The Cr and Cb components have a narrower
bandwidth because these are less sensitive to
human eye. The chrominance components are
usually decimated by two, both horizontally and
vertically resulting in a reduced number of
samples.

6. Spatial Redundancy
(Intra – frame compression)
 Within a single picture many blocks
have same value.
 Redundant
 DCT

7. Temporal redundancy
(Inter – frame compression)

8. Temporal redundancy…
(Inter – frame compression)
 How do we exploit this ?
 Send image differences
 Consecutive images are very similar.
 Difference images are spatially much more redundant
than real images.
 Exploit spatial redundancy of difference images!
 Motion vectors
 What if the camera moves?
 What if objects move?
 Use motion estimation before calculating the
difference image!

9. Entropy Redundancy
(Lossless compression)
 In picture sequence some values
occur very often
 Frequently appearing signal values
may be assigned a smaller length of
bits, thus eliminating considerable
amount of redundancy

10. Psycho-visual redundancy
 Human visual system
 Different sensitivity to different
information
 Human processing
 We only see some parts of the image
 Our brain completes the rest

11. Psycho-visual redundancy…
 Human sensitivity
 We notice errors in homogenous regions
 Low frequencies
 We notice errors in edges
 High frequencies
 We don’t notice noise in textured areas
 Medium frequencies

12. Image compression
 Lossy
 We do not obtain an exact copy of our
compressed data after decompression
 Very high compression rates
 Increased degradation with successive
compression / decompression
 Lossless
 We obtain an exact copy of our compressed
data after decompression
 Lower compression rates
 Freely compress / decompress images

13. Lossy Image Compression
 Acceptable for most real images and
situations.
 Very popular: JPEG
 We can control the level of
compression vs. Quality of the
resulting image.
 How do we do this?

14. Lossy Image Compression…

15. Discrete Cosine Transform (DCT)

16. Discrete Cosine Transform (DCT)

17. Video compression

18. Video compression
 Exploiting temporal redundancy
 Using all other redundancies for JPEG:
 Compression factor - 10:1
 Exploiting temporal redundancy for
MPEG-2:
 Compression factor – 100:1
 Temporal redundancy is of vital
importance to video compression!

19. Video compression…
 Goal: Exploit the temporal redundancy
 Predict current frame based on previously coded
frames
 Three types of coded frames:
 I-frame: Intra-coded frame, coded independently of all
other frames
 P-frame: Predicatively coded frame, coded based on
previously coded frame
 B-frame: Bi-directionally predicted frame, coded based on
both previous and future coded frames

20. Video compression…
 Simple frame differencing fails when there is
motion
 Must account for motion
 Motion-compensated (MC) prediction
 MC-prediction generally provides significant
improvements
 Questions:
 How can we estimate motion?
 How can we form MC-prediction?

21. Video compression…Motion Estimation
 Ideal situation:
 Partition video into moving objects
 Describe object motion
 Generally very difficult
 Practical approach: Block-Matching Motion Estimation
 Partition each frame into blocks, e.g. 16x16 pixels
 Describe motion of each block
 No object identification required
 Good, robust performance

22. Example Use of I-,P-,B-frames:
MPEG Group of Pictures (GOP)
 Arrows show prediction dependencies
between frames

23. MPEG: Motion Picture Experts Group
 MPEG-1 (1992)
 Compression for Storage
 1.5Mbps
 Frame-based Compression
 MPEG-2 (1994)
 Digital TV
 6.0 Mbps
 Frame-based Compression
 MPEG-4 (1998)
 Multimedia Applications, digital TV, synthetic graphics
 Lower bit rate
 Object based compression
 MPEG-7
 Multimedia Content Description Interface, XML-based
 MPEG-21
 Digital identification, IP rights management