MPEG 4 Codec Overview

1
MPEG 4 Codec Overview

• Study by
• Harsh Dhand

2
Contents

• Introduction
• Compression is required
• Systems Model
• Basics of Compression
• Advanced Features of MPEG4

3
MPEG4 is a Standard

• MPEG4 ISO/IEC Standard for coded representation
  of audio and visual data for transmission.
• Defines Coded representation method of
  decoding.
• Does not give implementation.

4
An Industry Breakthrough

• MPEG-4 set of technologies to satisfy the needs
  of authors, service providers and end users
  alike.
• Higher levels of interaction with content,
  multimedia to new networks, including those
  employing relatively low bitrate, and mobile
  ones.

5
Contents

• Introduction
• Compression is required
• Systems Model
• Basics of Compression
• Advanced Features of MPEG4

6
Why Compress/Encode
7
Compression MPEG4

• Compression (or encoding)
• key to enabling both wireless/wired transmission
  and storage of digital
• video content.
• MPEG4
• Offers extremely high image quality Compression
  in the rates of 701 to 2001.

8
DECoder
9
Contents

• Introduction
• Compression is required
• Systems Model
• Basics of Compression
• Advanced Features of MPEG4

10
Systems Model

• From Systems perspective can be thought
• of as layers or functional pieces in an
  architecture.
• Uses VRML for scene representation.
• BIFS Binary Format for Stream Representation

11
Layers

• Composition Layer Only in Decoder for
  reconstruction of scene.
• Media Layer Actual Encoding/Decoding
• Media representation in BIFS and ODs
• Sync Layer Tradeoffs of optimal bandwidth, right
  format and timely transmission.
• Transport Layer Media unaware delivery aware

12
(No Transcript)
13
Delivery of Streaming data

• DMIF (Delivery Multimedia Integration Framework)
  Grouping of elementary streams with low
  multiplexing overhead
• TransMux layer Transport services matching
  requested QoS
• Only Interface specified
• Can use any suitable Transport Protocol
• Essential Synch Layer
• Optional FlexMux Layer

14
(No Transcript)
15
Layers (contd)

• DMIF Interface between application transport.
• BIFS describes spatio-temporal arrangements of
  objects in scene.
• ODs define relationship between ESs pertinent to
  each object. URLs to access ESs, characteristics
  of Decoders, IP rights.

16
Contents

• Introduction
• Compression is required
• Systems Model
• Basics of Compression
• Advanced Features of MPEG4

17
Video Encoder Block Diagram
residual
Temporal model
Spatial model
Video input
coefficients
Entropy encoder
Encoded output
vectors
Stored Frames
18
Temporal Model

• Prediction from Previous Video Frame

19
(No Transcript)
20
Temporal Model

• Prediction from Previous Video Frame
• Changes due to motion Ideally estimate
  trajectory.optical flow
• Block Based Motion estimation Compensation
• Sub Pixel Motion Compensation

21
Macroblock Motion Compensated Prediction

• Macroblock 16 x 16 pixel region frame
• Estimation Finding sample region in reference
  frame closely matching current.
• Compensation Residual Macroblock motion
  vector(relative position)
• Issues
• Intra or Inter mode.
• Compensation Block size
• Sub Pixel Motion Compensation

22
(No Transcript)
23
Image/Spatial Model

• Predictive Image Coding

24
Transform Coding

• DCT Image Data represented in frequency domain,
  helps in Compression.
• Wavelet Transform Pair of filters to decompose
  into LF HF band.

25
Quantization

• Scalar Quantization Maps one sample of input
  signal to one quantized output.
• Vector quantizer Set of input Data to single
  codeword.

26

• DCT? Quantization ? Reordering ? RLE
• Reordering/Scan Zig Zag scan, either ideal as
  for frame for field (interlaced sampling) may be
  different.
• Run Level Encoding
• Input array 16,0,0,-3,5,6,0,0,0,-7,
• Output array (0,16),(2,-3),(0,5),(0,6),(4,-7)

27
(No Transcript)
28
Entropy Encoder

• Predictive Coding Object motion extending across
  large regions of frame. Motion Vector predicted
  on basis of neighbouring.
• MVD encoded and transmitted.
• Variable length Coding
• Huffman Coding
• Pre Calculated Huffman based
• Arithmetic Coding (Coz integral methods are
  sub optimal)

29
(No Transcript)
30
Contents

• Introduction
• Compression is required
• Systems Model
• Basics of Compression
• Advanced Features of MPEG4

31
Composition of Media Objects

• Media objects Compound and primitive.
• Based on VRML Structure functionality.
• Freedom to
• Place MOs anywhere
• Apply transforms
• Group to form complex MOs
• Change users viewing listening points

32
(No Transcript)
33
Video Objects

• Entity that a user is allowed to access and
  manipulate.
• Area of Video Scene occupying space and time.
• VOP Instance of VO at a particular point.
• Traditional
• Each VOP as single frame of video
• VO sequence of frames.

34
(No Transcript)
35
Definitions

• Tool Subset of coding functions to support
  specific feature (eg Basic Video coding,
  interlaced video,etc)
• Object Video Element, coded using one or more
  tool.
• Profile Set of object types that a CODEC is
  expected to handle.
• Levels Define constraints on parameters of
  bitstream.

36
Coding Rectangular Frames Simple Profile

• Based on DPCM/DCT Model.

37
(No Transcript)
38
Tools Used
39
VLE
VLD
40
Reconstructed Frame
ME
Source Frame
VLE
DCT
Q
RLE
ReOrder
MCP
Coded I-VOP
VLD
RLD
IDCT
Q(-1)
ReOrder
MCR
Decoded Frame
41
Short Header

• Compatibility between I-VOP/V-POP of MPEG4 and
  baseline mode of H.263
• Macroblocks within VOP organized in Groups of
  Blocks
• Each GoB may start with resynchronization marker

42
Coding Efficiency Tools

• When Short Header mode not enabled
• Four Motion Vectors per Block
• Option to chose a smaller motion compensation
  block size.
• Increased Overhead of 4 motion vectors
• Unrestricted Motion Vectors Motion Vector to
  point outside boundary of reference VOP

43
Transmission Efficiency Tools

• Video Packet Analogous to slice in MPEG1/2.
• Data Partitioning Reorganise Coded Data to
  reduce impact of transmission of errors.
• Reversible VLCs

44
Advanced Simple

• B-VOP
• Quarter-Pixel Motion Vectors
• Global Motion Compensation

45
Advanced Real Time Simple

• NEWPRED (multiple prediction references) used to
  limit temporal propagation of errors.
• Dynamic Resolution Conversion or Reduced
  resolution Mode To handle increased detail or
  rapid motion, over limited bitrate channel.

46
Coding Arbitrary Shaped Regions

• Shape Coding Alpha Blocks (16x16 pixel area)
  External, internal or crossing boundary.
• If crossing Shape information or transparency.
• Binary Scale (Core Profile)
• Grey Scale (Main profile)
• Motion Compensation Boundary Pixels of reference
  VOP padded to edges

47

• Advanced Features added on by
• Core Profile
• Main Profile
• Advanced Coding Efficiency Profile