1. CT20A6100 MACHINE VISION AND DIGITAL IMAGE ANALYSIS MPEG Pauli Jutila Cristina Petre

2. Contents 1. Introduction to MPEG 2. MPEG Family 3. Intra Frame Coding Techniques 4. Non-Intra Frame Coding Techniques 5.Summary

3. Why do we need video compression? Example video compression - 1920x1080 pixels 30 frames/sec 8 bits/color => 1.5Gb/sec - 6MHz channel bandwidth => 19.2Mb/sec => compression 83:1

4. The MPEG Family

5. MPEG-1 ● Basic form using motion compensated block-based transform coding techniques ● Optimized to work for 352x240 at 30frames/sec or 352x288 at 25frames/sec,possible to go up to 4095x4095 ● Defined for progressive frames only ● No direct provision for interlaced video applications

6. MPEG-2 ● Designed for digital television broadcasting ● Potentially high quality video (4-9Mb/sec) ● Profiles and levels : profile - bitstream scalability, colorspace resolution level - image resolution, maximum bit-rate/profile ● MP@ML : 720x480 at 30 frames/sec MP@ML ● MP@HL : 1920x1080 at 30 frame/sec MP@HL

7. MPEG-4 ● Interactive content ● Object-oriented composite files ● Image blocks are arbitrary shaped ● Multiplexes and synchronizes data ● Divided into a number of parts-compatibility issue

8. Coding Techniques

9. Intra Frame Coding Video FilterDCTQuantizer Run-Lenght VLC Bit-Stream Buffer Bit-Rate Control

10. Video Filter The human eye is less sensitive to chrominance than luminance. RGB  YCbCr Choice of sub-sampling ratios: 4:4:4 (Professional, Post-Production) 4:2:2 (same as above) 4:1:1 (DV Camcorder) 4:2:0 (Web, DVD)

11. Discrete Cosine Transform Nearly optimal in energy consentration and decorrelating The human eye is less sensitive to changes in higher frequencies Calculated in blocks, usually 8x8

12. DCT Coefficient Quantization 12-bit coefficient is divided by a corresponding value from intra quantization matrix Each value in the resultant matrix is pre-scaled by multiplying by a single value, known as the quantizer scale code (1-112) The goal is to force as many of the DCT coefficients to zero or near zero

13. Variable Length Coding Zigzag-scanning  long runs of consecutive zeros Alternate pattern in MPEG-2 (for interlaced video) Huffman coding

14. Coding Techniques

15. I,P,B Frames I-pictures: intra-coded no motion-compensated prediction used as reference picture P-picture: inter-coded forward predicted (from I or P frame) B-picture: inter-coded bi-directional prediction high coding efficiency

16. I,P,B Frames

17. Motion Estimation

18. Matching

19. Prediction Error Compensation

20. Non-Intra Decoding

21. Summary ● Video compression is needed! ● MPEG enables a wide range of applications ● Different standards but same principles ● Spatial and temporal processing

22. References ● Barry G. Haskell, Atul Puri, Arun N. Netravali, Digital Video: An Introduction to MPEG-2, Chapman and Hall, 1997. ● Joan L. Mitchell, William B. Pennebaker, Chad E. Fogg, Didier J. LeGall, MPEG Video Compression Standard, Chapman and Hall, 1997. ● John Wiseman, An Introduction to MPEG Video Compression, 1998 ● Marius Preda, Les standards MPEG dans l'espace multimedia, 2005 ● E. G. Richardson, Video Codec Design, Wiley, 2002 ● Wikipedia