BSc (Hons) Broadcast Technology
Unit BTC305: Media Compression
Handout Three: Interframe Compression
Lecturer: Martin Uren
Compression of Moving Images
We have already examined the issues surrounding the compression of
still images using JPEG and its subsequent adapting for video applications
under the generic description of Motion JPEG (M-JPEG). Now we need to
consider additional methods for the compression of video sequences.
MPEG are the Moving Picture Experts group, originally established to set
standards for the compression of feature films onto a CD format (called
Video-CD and using the MPEG-1 standard), but later given the broader
brief of compressing sound and pictures for a wide range of applications.
They subsequently developed the MPEG-2 and MPEG-4 compression
standards and the MPEG-7 standard for ‘metadata’ (the data describing
video and audio content).
MPEG deals with the compression of moving pictures and sound. In a
typical video sequence, much of the picture content is the same from
frame to frame, so MPEG devised techniques for only encoding the
difference between frames. This can involve motion prediction and
motion vectors to describe the movement of picture blocks from one
frame to the next. Their first standard, MPEG-1 resulted in a 1.2Mbit/s
video stream from a Video CD.
This is a huge reduction in data and gives a picture quality a little better
than VHS (most of the time!). The format was not commercially
successful, but is still used for example in news production systems to
give browse quality images on journalists’ workstations.
MPEG-2 from a DVD is typically between 4Mbit/s and 7Mbit/s, and the
resultant images are much higher quality than MPEG-1. When used in
transmission (for example DVB (Digital Video Broadcasting)) typical data
rates are between 2Mbit/s and 5Mbit/s.
From this we can roughly gauge the picture quality by knowing the bit
rate, but it’s not that simple. What factors will affect the viewed picture
quality if JPEG or MPEG encoding is used?
Video- H.261 Data Rate (bit/second)
10k 50k 100k 500k 1M 5M 10M 50M 100M 500M 1G 2G
Modem ADSL DTV SDI HDSDI
Compression System Quality and Bit-Rates
There are two basic techniques that are used:
1. Subtraction is used to find the differences between adjacent
frames. Only the differences are coded.
2. Motion compensated prediction takes a shifted block from the
previous frame to use as a prediction. A motion vector is generated
to describe the shift in position.
Motion Vectors as used in the MPEG1 standards are the result of
finding the best match for each macroblock in a picture using a Motion
Estimator. A macroblock is a 16 x 16 pixel group of 4 blocks. Motion
vectors are expressed in terms of vertical and horizontal displacement
from one field or frame to the next. Block matching is used to decide
where each macroblock has moved within a search area (perhaps 32 x
32 pixels). Each macroblock is therefore assigned a single motion vector.
Block matching is used within MPEG to produce predicted frames with
much higher levels of BRR than would otherwise be possible. The
technique is illustrated in the drawing below.
MotionVectors (or How to Turn a Daisy into a Golfball!)
Moving Picture Experts Group (originally Motion Picture Experts Group)
MPEG2 was set up by the ISO and the IEC in 1988 to standardise video
and audio compression coding methods. What are the difficulties to
consider when examining BRR methods for moving video images?
MPEG set out to code moving pictures and audio for use primarily with
compact disc formats. This work resulted in the MPEG-1 standard,
usable over a range of bit rates, but optimised for video data rates of
about 1.2Mbit/s (e.g. Philips Video-CD format). Note that MPEG-1
defines a multiplex of video, audio and data.
MPEG-1 reduces a BT.601 TV signal to SIF (Standard Image Format)
with 352 pixels x 288 lines (PAL) or 352 x 240 (NTSC). SIF also halves
the vertical colour resolution (i.e. 2:1:0 format) and ignores interlace.
This pre-processing reduces the data rate at the input to the main
MPEG have defined a process that applies motion compensated
interframe prediction to the JPEG digital image coder shown above. A
shifted macroblock of four 8 x 8 pixel blocks from the previous frame is
used as a prediction.
A motion vector is generated to describe the shift in position of each
macroblock in the frame. These motion vectors are encoded with the
picture data in the MPEG-1 bitstream. The resulting encoder and
decoder block diagrams are shown overleaf. MPEG-1 uses bi-directional
prediction for some frames using past and future frames. How many
motion vectors are there for each frame?
Moving Picture Experts Group, also known as Motion Picture Experts Group
MPEG set goals for the picture quality likely to be obtained from a
compressed digital capacity, as shown in the table below. MPEG-1 was
optimised for about 1.2Mbit/s, while MPEG-2 addresses higher quality
moving images and is optimised for data rates of 2Mbit/s to 15Mbit/s.
MPEG-2 also handles flexible, dynamically variable data rates, makes use
of the temporal redundancy in interlaced pictures, and is designed to be
fully compatible with the data rates used in telecomms networks. It is
backwards compatible with MPEG-1.
MPEG-2 Video Compression
Source coding for the digital video follows the scheme shown above for
MPEG-1. It results in an Elementary Stream. MPEG defined a family of
sub-systems described in terms of Profiles and Levels, as shown in the
table overleaf. MPEG-2 MP@ML (Main Profile at Main Level) has been
almost universally used for digital transmission systems. However, it is
not high enough quality for use in studios, so a further level was added.
MPEG-2 was aimed at digital distribution and transmission systems. Much
interest was expressed in using MPEG-2 in the production chain. This is
where MPEG-2 MP@ML has some limitations:
1. 4:2:0 sampling (for 625 lines - 4:1:1 for 525 line systems), rather
than the 4:2:2 sampling of BT.601. This reduces the colour
resolution, most critically for chromakey applications.
2. Maximum data rate of 15Mbit/s. This is not adequate for contribution
quality or use in a studio centre.
3. Only the active picture area is coded.
These problems led to the establishment of a 422 sub-group of MPEG
who developed the 422-profile (MPEG-2 4:2:2P@ML).
Like MPEG-1, MPEG-2 uses bi-directional prediction for some frames
using past and future frames. This makes the encoding and decoding
more complex than JPEG, and also leads to difficulties in a studio centre
with switching and editing.
Three types of compressed frames are used:
I = Intra coded picture
P = Predictively coded picture
B = Bidirectionally coded picture
Frames are combined in a sequence, or GOP3. An example of a 12-frame
GOP is shown overleaf.
Betacam SX is a proprietary frame-accurate editable variant of MPEG-
2. It is an application of MPEG-2 422P@ML using 4:2:2 sampling and a
10:1 compression ratio. The data rate covers the range 15Mbit/s to
50Mbit/s. It also codes up to 608 lines, to include some field blanking
lines, and uses a 2-frame GOP of the form I B I B.
Group Of Pictures
Quality Digital Equivalen
1250 line HDTV with no perceptible degradation 40 Mbit/s
HDTV quality with some distortion on critical scenes 20 Mbit/s
625 line studio quality with no perceptible degradation 10 Mbit/s BT.601
625 line quality with some distortion on critical scenes 5 Mbit/s PAL
Reduced quality (312 line) with no perceptible degradation 2.5 Mbit/s
Reduced quality (312 line) with some distortion on critical 1.2 Mbit/s VHS
MPEG Compressed Picture Quality
Simple Main Scalable Scalable High
No B Frames B Frames B Frames B Frames B Frames
Levels 4:2:0 4:2:0 4:2:0 4:2:0 4:2:0 & 4:2:2
1920 x 1152 80Mbit/s 100Mbit/s
1440 x 1152 60Mbit/s 60Mbit/s 80Mbit/s
720 x 576 15Mbit/s 15Mbit/s 15Mbit/s 20Mbit/s
352 x 288 4Mbit/s 4Mbit/s
MPEG Family Members