The Internet Encyclopedia (Volume 3)

P1: IML/FFX P2: IML/FFX QC: IML/FFX T1: IML

Video ̇Compression ̇OLE WL040/Bidgolio-Vol I WL040-Sample.cls September 14, 2003 18:10 Char Count= 0

DIGITALVIDEOSIGNALCOMPRESSION 543

researchers compare the effectiveness of video compres-

sion algorithms by plotting empirically determined qual-

ity measures as a function of observed bit rate for a given

set of test sequences. Almost invariably, the observed video

quality improves with increased bit rate.

Principles of Digital Video Compression

Digital video compression operates by eliminating re-

dundant spatial, temporal, hyperspectral, statistical, or

pyschovisual information. A brief inspection of Figure

1a shows considerable overlap between frames 1232 and

1233 but a discontinuous change of scene (likely as a re-

sult of editing) between frames 1231 and 1232. All the

frames shown exhibit strong local spatial correlation, the

motion of the astronauts exhibits temporal correlation,

the viewer’s attention is focused on the motion of astro-

nauts and the expressions on their faces, and the space suit

color contrasts well with the background. Figure 1b illus-

trates the well-known video artifact of “dropped frames”

when this movie is compressed at an average data rate

less than the original 293 kbps.

Human Destination

Compression of video sequences intended for human

viewing may take place by a combination of lossy and

lossless compression steps. Local spatial correlation may

be removed by intraframe coding techniques such as

block matching or transform coding followed by quan-

tization of the transform coefficients (see Data Com-

pression).

Temporal correlation may be removed by interframe

coding techniques predicting the motion vectors ob-

served in differences between successive frames. Spectral

correlation may be removed by applying temporal decor-

relation or modeling techniques to the spectral dimen-

sions of the hyperspectral imagery; coding redundancy

may be removed through careful design of compression

codes, while psychovisual redundancy may be addressed

by techniques that drop frames and increase the bit rate

in regions containing human faces, sharply contrasting

regions and trajectories of distracting objects that move

rapidly through the peripheral field of view.

Machine Destination

In scientific and medical applications, imagery may be

correlated in any of the preceding ways. There are likely to

be additional constraints on the compression which may

include fidelity criteria (such as 90% of the encircled en-

ergy to remain within the pixel for remotely sensed land

use data, or the peak difference in value reconstructed

from the compressed imagery to be within 10% of the orig-

inal value for 90% of the reconstructed pixels) together

with processing constraints such as the bit rate must re-

main approximately constant within the communication

channel capacity regardless of video quality or content

and consume no more than 5% of the available process-

ing resource.

As an example, Freedman, Boggess, and Seiler (1993)

and Freedman and Farrelle (1996) set forth criteria for

the experimental compression system developed to opti-

mize real-time calculations on large diverse data sets from

NASA’s Cosmic Background Explorer mission in a clus-

tered computing environment in which the application

software exceeded 1M lines of code:

Provide compression transparently without changing the

application software.

Compress instrument pipeline and science analysis data

products to better than 16–50%.

Process compressed data at a worst case throughput not

less than 90% of uncompressed data processing.

Preserve required accuracy of instrument housekeeping

and scientific data according to specified validation

criteria.

Exceed bitwise reliability of 10–^13 on average (flawless

compression of 380 GB). Several times this factor is

desirable.

Support full random access to data records.

Provide a capability to select specific classes of data for

compression.

Provide a capability to select a compression scheme (rep-

resentation) for each field of a data record.

Preserve overlaps in separately processed data segments.

Store search keys (e.g., time code and pixel address) in

plain codes.

Optimize choice of representation, combining a priori sci-

entific knowledge with adaptive knowledge of data.

Provide mechanisms for easy user configuration and

database management of compressed data.

Pre- and Postprocessing

Video source material is often pre-processed before en-

coding and post-processed before distribution. Com-

mon preprocessing steps include digital nonlinear editing

(Ohanian, 1998), transcoding from another compressed

representation, and the conversion of film-originated ma-

terial from 24 to 29.97 fps via the telecine process just

discussed. Common postprocessing steps include digital

nonlinear editing, the inverse telecine process, and the

manual deletion or recoding of specific frames.

Motion Estimation and Compensation

Strong temporal correlation between successive video

frames suggest that the bit rate of a video sequence may

be reduced by interframe coding methods such as condi-

tional replenishment, motion-compensated coding, and

three-dimensional transform coding.

The conditional replenishment approach seeks to en-

code only the difference in pixel values which change be-

tween successive frames and requires considerably less

processing power and simpler algorithms than either

motion-compensated coding or three-dimensional trans-

form coding. Motion-compensated methods usually re-

sult in data rates below that resulting from conditional

replenishment and comprise determining motion vectors

from frames preceding the current frame by methods

such as block matching, region matching, or analysis of

optical flow, predicting the motion vectors for succes-

sive frames, estimating the prediction error of these mo-

tion vectors, and encoding the predicted motion vectors