Video Compression 2:

Bi-directional Coding
Multimedia Systems (Module 4 Lesson 3)

Summary: Sources:
r MPEG Coding r “Digital Compression for
Multimedia: Principles and
m Bi-directional Motion-
Standards”, Jerry D.
Compensation
Gibson, Toby Berger, Tom
r MPEG Parameters Lookabaugh, Dave
r MPEG 2 and 4 Lindbergh and Richard L.
Baker.
r My research notes

1
MPEG
r "Moving Picture Coding Experts Group", standards body for
delivery of video and audio.
r MPEG-1 Target: VHS quality on a CD-ROM or Video CD
(VCD) (352 x 240 + CD audio @ 1.5 Mbits/sec)
r MPEG-2 allows different levels and profiles
r Both standards have four parts:
m Video: Defines the video compression decoder
m Audio: Defines the audio compression decoder
m System: Describes how various streams(video, audio or generic
data) are multiplexed and synchronized.
m Conformance: Defines a set of tests designed to aid in
establishing that particular implementations conform to the
design.

2
The Problem
r Some macroblocks need information that is not present in
the previous reference frame.
m Maybe, such information is available in a succeeding frame!

Previous Frame Current Frame Next Frame

r Add a third frame type (B-frame): To form a B-frame,

search for matching macroblocks in both past and future
frames.
r Typical pattern is IBBPBBPBB IBBPBBPBB IBBPBBPBB
Actual pattern is up to encoder, and need not be regular.
3
Bitstream order vs. Display order
Bitstream (Transmit) order:
1(I), 4(P), 2(B), 3(B), 7(P), 5(B), 6(B), 10(I), 8(B), 9(B)

P B B I B B P B B P B B I B

-3 -2 -1 1 2 3 4 5 6 7 8 9 10 11

4
Frame and Macroblock Prediction Types
Some definitions:
m Anchor frame: A frame that can be used for prediction
We now discuss the various frame types and Macroblock types

Macroblock Type Prediction

Nonpredicted Macroblock none
Backward-predicted macroblock References temporally nearest subsequent
anchor frame
Forward-predicted macroblock References temporally nearest previous
anchor frame
Bidirectionally predicted macroblock Averages predictions from temporally nearest,
previous and subsequent anchor frames

Frame Type Anchor Frame Macroblock Types

I-frame Yes Nonpredicted

P-frame Yes Nonpredicted, Forward predicted

B-frame No Nonpredicted, Forward predicted,

Backward predicted, Bidirectionally predicted

5
Bidirectional Prediction
r B-frames allow effective prediction of uncovered
background, areas of the current picture that were not
visible in the past and visible in the future.
r B-frames can provide for interpolation equivalent to an even
finer degree than half-pixel (1/4 pixel for example).
r If good prediction is available in both the previous and
subsequent anchor frames, then averaging the two
predictors reduces noise and hence increases efficiency.
r -ve: Motion estimation becomes more complex (look farther).
r A ratio of 5:3:1 between the number of bits spent on I, P
and B frames is quite common.
r Errors in B-frames tend to limit the effect to that B-frame
only.

6
MPEG Notation

Though the standard does not dictate this, the pattern (order)
of frames are commonly referred by the following notation:
(N,M) where:
r N is the number of frames from one I-frame(inclusive)
to the next (exclusive).
r M is the number of frames from one anchor(inclusive) to
the next(exclusive).
The example sequence discussed before would be a (N=9, M=3 )
pattern.

7
Differences from H.261
r Larger gaps between I and P
frames, so need to expand motion
vector search range. Uni-Quant
for P and Non-uniform-Quant for
I.
r To get better encoding, allow
motion vectors to be specified to
fraction of a pixel (1/2 pixel).
r Bitstream syntax allows random
access, forward/backward play,
etc.
r Added notion of slice for
synchronization after loss/corrupt
data (see figure at right: 7 slices
in frame).

8
MPEG-1 Parameter Constraints

Parameter Constraint

Horizontal size < 720 pixels

Veritcal size < 576 lines

Total # of macroblocks per < 396

picture
Total # of macroblocks per < 396 x 25 (or 396 x 30)
second
Frame rate < 30

Bit rate < 1.86 Mbps

Decode buffer < 376832 bits

9
MPEG2
r Unlike MPEG-1 which is basically a standard for storing and
playing video on a single computer at low bit-rates, MPEG-2
is a standard for digital TV. It meets the requirements for
HDTV and DVD (Digital Video/Versatile Disc).
r MPEG2 Supports the following levels:
Level Size Pixels/sec Bit-rate(Mbps) Application
Low 352 x 288 x 30 3M 4 consumer tape equiv.
Main 720 x 576 x 30 12 M 15 Studio TV
High 1440 1440 x 1152 x 60 96 M 60 Consumer HDTV
High 1920 x 1152 x 60 128 M 80 Film Production

r It supports multiple profiles based on scalability

r Supports both field prediction and frame prediction.
r Besides 4:2:0, also allows 4:2:2 and 4:4:4 chroma-
subsampling

10
MPEG4
r Originally targeted at very low bit-rate communication (4.8
to 64 Kb/sec), it now aims at the following ranges of bit-
rates:
m video -- 5 Kb to 5 Mb per second
m audio -- 2 Kb to 64 Kb per second
r It emphasizes the concept of Visual Objects --> Video
Object Plane (VOP)
r Objects can be of arbitrary shape, VOPs can be non-
overlapped or overlapped
r Supports content-based scalability
r Supports object-based interactivity
r Individual audio channels can be associated with objects
r Good for video composition, segmentation, and compression;
networked VRML, audiovisual communication systems (e.g.,
text-to-speech interface, facial animation), etc.
r Standards being developed for shape coding, motion coding,
texture coding, etc.
11