MPEG-1 and MPEG-2 Digital Video Coding Standards
Author: Thomas SikoraPresenter: Chaojun Liang
Overview
• Introduction• Intra-picture Coding Techniques• Inter-picture Coding Techniques
What is MPEG?
• Stands for Moving Picture Expert Group
• Nickname to a family of International standards used for coding audio-visual information in a digital compressed format.
• Includes MPEG-1, MPEG-2, and upcoming MPEG-4.
MPEG-1
• Finalized in 1991• Optimized to work at 352*240, 30
f/s or 352*288, 25f/s• Bit-rate is optimized for application
of around 1.5 Mb/sec• Defined for progressive frames
only
MPEG-2• Finalized in 1994• Digital television broadcasting• Target bit-rate raised to 4-9Mb/sec• Main Profile, Main Level
– 720*480 pixels, 30 frame/sec, 15Mb/sec
• Main Profile, High Level (HDTV)– 1920*1280 pixels, 30 frame/sec,
80Mb/sec
MPEG Video Layers
Intra Frame Coding Techniques
Video Filter
DCT QuantizerRun-length Encoding
Bitstream Buffer
Bit-Rate Control
Video Filter
• Subjective Redundancy– Eyes more sensitive to changes in
Luminance, less to variations in Chrominance
• R G B -> Y Cb Cr• Sample Chrominance 2:1 in both
directions
R G B Y Cb Cr Y Cb Cr
Color Space Transform Sample
Video Filter
Statistical Redundancy
Discrete Cosine Transform
• Closely related to Discrete Fourier Transform
• Performed on 8*8 blockForward DCT
Inverse DCT
DCT Basis Patterns
DCT Basis Patterns
Increasing Vertical Frequency (Down)
Increase Horizontal Frequency (Right)
Variance Distribution of DCT Coefficients
DCT Coefficient Quantization
• Higher frequency -> Larger step size -> more coarsely quantized
• Force as many of the DCT-Coefficients as possible to zero or near zero
)),(
),((),(
vuQ
vuFndIntegerRouvuFQ
Example Coefficients
Run-Length Encoding
Run-Amplitude Pairs Bit Stream
Variable Length Code Table
Inter Frame Coding Techniques• Exploit temporal redundancy• Use Motion Estimation to eliminate
temporal redundancy
Picture Types
• Intra Pictures (I-Pictures)• Predicted Pictures (P-Pictures)• Bi-directional Pictures (B-Pictures)
Intra Pictures
• Coded with only info in the picture itself
• Random access points into the compressed data
• Moderate compress: about 2 bits per coded pixel
Predicted Pictures
• Coded with respect to the nearest previous I-Picture or P-Picture
Bi-directional Pictures
• Use both a past and future picture as a reference.
Video Stream Composition
• Encoder chose frequency and location of I-Pictures
• Typical display order
Video Stream vs Display order
Motion Estimation
Tree moved down to the right
People moved farther to the right than tree
Frame 1 Frame 2
Motion Estimation
• To adequately represent the change, or the difference, between 2 video frames
• Frame-wise direct minus not effective
• Macroblock-wise minus • 2-D spatial search for each
luminance macroblock
Macroblock Matching
Frame 1 Frame 2
Picture 1
Predicted Picture
Picture 2
Residual Error Picture
Coding of Residual Errors
• Similar to Intra frame coding• Different quantization matrix• Motion vectors are also coded
