|How the MPEG Process Works
compression dramatically decreases the amount of storage space required
to record motion picture sequences by eliminating redundant and non-essential
image information from the stored data. Less total bits means that motion
pictures can be transferred more rapidly, so rapidly in fact, that slow,
but inexpensive, communication lines and storage devices have entirely
new moving picture applications. The following set of pictures are actual
screen captures of a frame undergoing MPEG compression using the PixelTools'
frames in a motion picture sequence are usually very nearly identical.
Often the only difference is that some parts of the picture have shifted
slightly between the frames. MPEG compression exploits this temporal redundancy
by carving each new frame into convenient pieces and searching the previous
frame to determine where each piece came from. If the content of the current
frame was mostly sent in the previous frame, why send it all again? Just
send the instructions for shifting pieces of the previous frame to their
new positions in the current frame.
Within a single frame many patches, such as regions of sky or walls are
almost entirely the same color. Again, MPEG compression exploits this
spatial redundancy by carving images into convenient pieces and reducing
such patches to a single color. If several pixel points in the same area
are all very nearly the same color, why send the same color over and over?
Just send the color for the whole area once.
The human eye is forgiving of approximation and outright elimination of
the finest details in images. This is fortunate because there is proportionally
a lot more fine detail than coarse detail in images. MPEG compression
approximates the intensity of fine detail with just a few shades (and
progressively more shades for coarser detail) saving many bits over full
representation. The eye also sees less color changes per inch than it
does brightness variations. MPEG compression exploits this reduction by
eliminating non- essential color detail that the human eye tends to gloss
image represents an image that has been carved up into about 300
squares to perform the compressions mentioned above. This image
is part of a well known test motion picture sequence that depicts
passage from left to right along a row of houses fronted by a strip
of colorful flowers. The gray patches represent areas that were
found in the previous frame and shifted to new positions to form
this frame. A uniform gray patch indicates an exactly matching area
was found. Mottled gray patches indicate a near match. This operation
is performed to compress temporal redundancy, the nearly complete
replication of the same picture in two adjacent frames of a sequence.
that close matches for some squares could not be found in the previous
frame. There was no close match for the new image detail that shifted
into view on the right edge or for the tree that is passing by rapidly
in the foreground. The process of finding the squares in another picture
is called motion compensation and is very computationally intensive. Applying
motion compensation to reconstruct an compressed MPEG sequence is, by
contrast, a much easier task. This characteristic is called asymmetrical
compression or encoding: hard to compress but easy to decompress or decode.
blowup of a section of Image 1 with the blue lines
(called motion vectors) indicating the exact location of the best
matching group of pixels in the previous frame used in motion compensation.
This motion vector (maybe 4 bits of data) is recorded in the MPEG
stream instead of the group of pixels (2048 bits).
original picture with a tremendous amount of detail and color has been
transformed to the representation shown in Image 1 with
the first compression step. It is important to notice that resultant picture
is now nearly one uniform color (the gray) with some finely mottled detail.
Represents Image 1 after
preparation for the second compression step. Picture
3 is measure of the amount of detail remaining in the picture
after the first step. More and brighter areas represent the uncompressed
detail remaining in the gray picture. Note that this image is mostly
black, meaning that little detail remains. The detail which does
remain is typically concentrated into one corner of the original
square patches which indicates that it is mostly of a single color.
The final compression step consists of reducing the
number of shades used to represent the fine picture detail. This
corresponds to reducing even further the intensity of the light
areas representing fine picture detail (the areas not in the single
color corners.) In addition to applying more reduction to fine detail
within a single square, MPEG provides separate weight for each whole
square. Squares containing more fine detail are compressed relatively
more. Image 4 shows the relative amounts that each
square patch is reduced.
final representation is nearly all black which is nearly all zeros or
very small (short) numbers. These short numbers require very few bits
to be stored or transmitted. The formerly, complex and colorful picture
sequences have been reduced to the minimum essential representation required
for storage or transmission in highly compressed MPEG format.
2002 PixelTools Corporation. All rights reserved. Revised © 2009.