MULTIMEDIA DATABASES

Images, sounds and movies can be stored, retrieved and played by many databases. In future,
multimedia databases will bcome a main source of interaction between users and multimedia
elements. Multimedia storage and retrieval Multimedia storage is characterized by a number
of considerations. They are:

(i) massive storage volumes

(ii) large object sizes

(iii) multiple related objects

(iv) temporal requirements for retrieval

Massive Data Volumes

A single multimedia document may be a combination of different media Hence indexing of documents,
fi lms and tapes is more complex. Locating massive data volumes requires searching through massive
storage files.

Locating and indexing systems can be understood only by a few key staff personnel. Hence it requires a
major organizational eff0l1 to ensure that they are returned in proper sequence to their original storage
location.

storage technologies

There are two major mass storage technologies used currently for storage of multimedia documents.
(i) Optical disk storage systems. (ii) High-speed magnetic storage.

Advantages of Optical disk storage systems:

(i) Managing a few optical disk platters in a juke box is much simpler than man;lging a large magnetic
disk farm. (ii) Optical disk storage is excellent storage system for off line archival of old and infrequently
referenced documents for significant periods of time

Multimedia object storage

Multimedia object storage in an optical medium serves its original purpose, only if it can be located fast
and automatically. A key issue here is random keyed Access t6 various components of hypermedia
database record. Optical media providcs very dense storage. Speed of retrieval is another consideration.
Retrieval speed is a direct result of the storage latency, size of the data relative to display resolution,
transmission media and speed, and decompression efficiency. Indexing is important for fast retrieval of
information. Indexing can be at multiple levels.

Multimedia document retrieval

The simplest form of identifying a multimedia document is by storage platter identification and its
relative position on the platter (file number). These objects can then be grouped using a database in
folders (replicating the concept of paper storage in file folders) or within complex objects representing
hypermedia documents.

The capability to access objects using identifiers stored in a database requires capability in the database
to perform the required multimedia object directory functions. Another important application for sound
and full motion video is the ability to clip parts of it and combine them with another set.

Indexing of sound and full-motion video is the subject of intense debate and a number of approaches
have been used.

Database Management Systems for Multimedia Systems

Since most multimedia applications are based primarily on communications technologies, such as
electronic mail, the database system must be fully distributed. A number of database storge choices are
available.

The choices available are:

* Extending the existing relational database management systems, (RDBMSs) to support the
various objects for multimedia as binary objects.

* Extending RDBMSs beyond basis binary objects to the concepts of inheritance and classes.
RDBMSs supporting these . features provide extensions for object-programming front ends and/or C++
support.
* Converting to a full fledged object oriented database that supports the standard SQL
language.

* Converting the database and the application to an objectoriented database and using an
object-oriented language, or an object-enabled SQL for development.

Multimedia applications combine numerical and textual data, graphics from GUI front-ends, CAD/CAM
systems and GIS applications, still video, audio and full-motion video with recorded audio and annotated
voice components. Relational databases, the dominent database paradigm, have lacked the ability to
support multimedia databases. Key limitations of relational database systems for implementing
multimedia applications stem from two areas: the relational data model and the relatIonal
computational model.

RDBMSs have been designed to manage only tabular alphanumeric forms of data (along with some
additional data types stored in binary form such as dates).

RDBMS EXTENSIONS FOR MULTIMEDIA

Binary Large Object (BLOB) is a data type which has been adapted by most of the leading relational
databases. BLOBs are used for objects such as images or other binary data types.

The relational database is extended to access these BLOBs to present the user 'with a complete' data
set. Extended relational databases provide a gradual migration path to a more object-oriented
environment. Relational database tables include location information for the BLOBs which may be
stored outside the database on separate image or video servers. Relational databases have the strength
of rigorous set management for maintaining the integrity of the database

Object-Oriented Databases for Multimedia

.In object databases, data remains in RMS or flat files. Object databases can provide the fastest route to
multimedia support. Object programming embodies the principles of reusable code and modularity. This
will ease future maintenance of these databases.
Object database capabilities such as message passing, extensibility, and the support of hierarchical
structures, are important for multimedia systems.

We can develop the application fastest class definitions. ODBMSs are extensible. They allow incremental
changes to the database applications.

Extensibility: Extensibility .means that the set of operations, structures and constraints that
are available to operations are not fixed, and developers can define new operations, which can then be
added as needed to their application.

Object-oriented software technology has three important concepts. They are:

Encapsulation: It is the ability to deal with software entities as units that interact in pre-defined
and controllable manner, and where the control routines are integral with entity.

Association: It is the ability to define a software entity in terms of its di fferences from another
entity. Classification: It is the ability to represent with a single software entity a number of data items
that all have the same behavior and the same state attributes.

Object orientation helps to organize the software in a more, modular and re-usable
manner. Encapsulation allows for the development of open systems where one part of the application
does not need to know the functioning of other part. It also provides autonomy; Autonomy means we
can interface to a variety of external programs can be built in one class of objects and the storage of the
data in another class of objects.

Database Organization for Multimedia Applications

Data organization for multimedia systems has some key issues. They are:

(l) Data independence (2) Common distributed database architecture

(3)Distributed database servers· (4) Multimedia object management.

Data Independence

Flexible access by a number of databases requires that the data be independent from the application so
that future applications can access the data without constraints related to a previous application.

Key features of data independent designs are:

1.Storage design in independent of specific applications.

2.Explicit data definitions are independent of application program.

3.Users need not know data formats or physical storage structures.

4.Integrity assurance in independent of application programs.

5.Recovery in independent of application programs.

Distributed Data servers

Distributed database servers are a dedicated resource on a network accessible to a number of

applications. The database server is built for growth and enhancement, and the network provides the
opportunity for the growth of applications and distributed access to the data.

Multimedia Object Management

The object management system must be capable of indexing, grouping and storing multimedia objects
in distributed hierarchical optional storage systems, and accessing these objects on or keyed basis.

The design of the object management system should be capable indexing objects in such a manner that
there is no need to maintain multiple storage copies.
Multimedia transactions are very complex transactions. We define a multimedia transaction as the
sequence of events that starts when a user makes a request to display, edit, or print a hyper media
document. The transaction is complete when the user releases the hypermedia document and stores
back the edited versions or discards the copy in memory (including virtual memory) or local storage ..

COMPRESSION AND DECOMPRESSION

Compression is the way of making files to take up less space. In multimedia systems, in order to manage
large multimedia data objects efficiently, these data objects need to be compressed to reduce the file
size for storage of these objects.

Compression tries to eliminate redundancies in the pattern of data.

For example, if a black pixel is followed by 20 white pixels, there is no need to store all 20 white pixels. A
coding mechanism can be used so that only the count of the white pixels is stored. Once such
redundancies are removed, the data object requires less time for transmission over a network. This in
turn significantly reduces storage and transmission costs.

TYPES OF COMPRESSION

Compression and decompression techniques are utilized for a number of applications, such as facsimile
system, printer systems, document storage and retrieval systems, video teleconferencing systems, and
electronic multimedia messaging systems. An important standardization of compression algorithm was
achieved by the CCITT when it specified Group 2 compression for facsimile system. .

When information is compressed, the redundancies are removed.

Sometimes removing redundancies is not sufficient to reduce the size of the data object to manageable
levels. In such cases, some real information is also removed. The primary criterion is that removal of the
real information should not perceptly affect the quality of the result. In the case of video, compression
causes some information to be lost; some information at a delete level is considered not essential for a
reasonable reproduction of the scene. This type of compression is called lossy compression. Audio
compression, on the other hand, is not lossy. It is called lossless compression.
Lossless Compression.

In lossless compression, data is not altered or lost in the process of compression or decompression.
Decompression generates an exact replica ofthe original object. Text compression is a good example of
lossless compression. The repetitive nature of text, sound and graphic images allows replacement of
repeated strings of characters or bits by codes. Lossless compression techniques are good for text data
and for repetitive data in images all like binary images and gray-scale images.

Some of the commonly accepted lossless standards are given below:

v Packpits encoding (Run-length encoding)

v CCITT Group 3 I D

v CCITT Group 3 2D

v CCITT Group 4

v Lempe l-Ziv and Welch algorithm LZW.

Lossy compression is that some loss would occur while compressing information objects.

Lossy compression is used for compressing audio, gray-scale or color images, and video objects in which
absolute data accuracy is not necessary.

The idea behind the lossy compression is that, the human eye fills in the missing information in the case
of video.
But, an important consideration is how much information can be lost so that the result should not affect.
For example, in a grayscale image, if several bits are missing, the information is still perceived in an
acceptable manner as the eye fills in the gaps in the shading gradient.

Lossy compression is applicable in medical screening systems, video tele-conferencing, and multimedia
electronic messaging systems.

Lossy compressions techniques can be used alone o~' in combination with other compression methods
in a multimedia object consisting of audio, color images, and video as well as other specialized data
types.

The following lists some of the lossy compression mechanisms:

ü Joint Photographic Experts Group (JPEG)

ü Moving Picture Experts Group (MPEG)

ü Intel DVI

ü CCITT H.261 (P * 24) Video Coding Algorithm

ü Fractals.

Compression schemes are

Binary Image compression schemes

Binary Image Compression Scheme is a scheme by which a binary image containing black and white pixel
is generated when a document is scanned in a binary mode.
The schemes are used primarily for documents that do not contain any continuous-tone information or
where the continuous-tone information can be captured in a black and white mode to serve the desired
purpose.

The schemes are applicable in office/business documents, handwritten text, line graphics, engineering
drawings, and so on. Let us view the scanning process.A scanner scans a document as sequential scan
lines, starting from the top of the page.

A scan line is complete line of pixels, of height equal to one pixel, running across the page. It scans the
first line of pixels (Scan Line), then scans second "line, and works its way up to the last scan line of the
page. Each scan line is scanned from left to right of the page generating black and white pixels for that
scan line.

This uncompressed image consists of a single bit per pixel containing black and white pixels. Binary 1
represents a black pixel, binary 0 a white pixel. Several schemes have been standardized and used to
achieve various levels of compressions. Let us review the more commonly used schemes.

1 .Packpits Encoding( Run-Length Encoding)

It is a scheme in which a consecutive repeated string of characters is replaced by two bytes. It is the
simple, earliest of the data compression scheme developed. It need not to have a standard. It is used to
compress black and white (binary) images. Among two bytes which are being replaced, the first byte
contains a number representing the number of times the character is repeated, and the second byte
contains the character itself.

In some cases, one byte is used to represent the pixel value, and the other seven bits to represents the
run length.

2. CCITT Group 3 1-D Compression

This scheme is based on run-length encoding and assumes that a typical scanline has long runs of the
same color.

This scheme was designed for black and white images only, not for gray scale or color images. The
primary application of this scheme is in facsimile and early document imaging system.

Huffman Encoding

A modified version of run-length encoding is Huffman encoding.

It is used for many software based document imaging systems. It is used for encoding the pixel run
length in CCITT Group 3 1-dGroup 4.

It is variable-length encoding. It generates the shortest code for frequently occurring run lengths and
longer code for less frequently occurring run lengths.

Mathematical Algorithm for huffman encoding:

Huffman encoding scheme is based on a coding tree.

It is constructed based on the probability of occurrence of white pixels or black pixels in the run length
or bit stream.

Table below shows the CCITT Group 3 tables showing codes or white run lengths and black run lengths.
For example, from Table 2, the run-length code of 16 white pixels is 101010, and of 16 black pixels
0000010111. Statistically, the occurrence of 16 white pixels is more frequent than the occurrence of 16
black pixels. Hence, the code generated for 16 white pixels is much shorter. This allows for quicker
decoding. For this example, the tree structure could be constructed.
The codes greater than a string of 1792 pixels are identical for black and white pixels. A new code
indicates reversal of color, that is, the pixel Color code is relative to the color of the previous pixel
sequence.

Table 3 shows the codes for pixel sequences larger than 1792 pixels.

CCITT Group 3 compression utilizes Huffman coding to generate a set of make-up codes and a set of
terminating codes for a given bit stream. Make-up codes are used to represent run length in multiples of
64 pixels. Terminating codes are used to represent run lengths of less than 64 pixels.
As shown in Table 2; run-length codes for black pixels are different from the run-length codes for white
pixels. For example, the run-length code for 64 white pixels is 11011. The run length code for 64 black
pixels is 0000001111. Consequently, the run length of 132 white pixels is encoded by the following two
codes:

Makeup code for 128 white pixels - 10010 Terminating code for 4 white pixels - 1011

The compressed bit stream for 132 white pixels is 100101011, a total of nine bits. Therefore the
compression ratio is 14, the ratio between the total number of bits (132) divided by the number of bits
used to code them (9).

CCITT Group 3 uses a very simple data format. This consists of sequential blocks of data for each
scanline, as shown in Table 4.

Coding tree for 16 white pixels

Note that the file is terminated by a number of EOLs (End of. Line) if there is no change in the line [rom
the previous line (for example, white space).
TABLE 4: CCITT Group 3 1D File Format

Advantages of CCITT Group 3 ID

CCITT Group 3 compression has been used extensively due to the following two advantages: It is simple
to implement in both hardware and software .

It is a worldwide standard for facsimile which is accepted for document imaging application. This allows
document imaging applications to incorporate fax documents easily.

CCITT group 3 compressions utilizes Huffman coding to generate a set of make-up codes and a set of
terminating codes for a give bit stream.

CCITT Group 3 uses a very simply data format. This consists of sequential blocks of data for each
scanline.

CCITT Group 3 2D Compression

It is also known as modified run length encoding. It is used for software based imaging system and
facsimile.

It is easier to decompress in software than CCITT Group 4. The CCITT Group 3 2D scheme uses a "k"
factor where the

image is divided into several group of k lines. This scheme is based on the statistical nature of images;
the image data across the adjacent scanline is redundant.
If black and white transition occurs on a given scanline, chances are the same transition will occur within
+ or - 3 pixels in the next scanline.

Necessity of k factor·

When CCITT Group 3 2D compression is used, the algorithm embeds Group 3 1 D coding between every
k groups of Group 3 2D coding, allowing the Group 3 1 D coding to be the synchronizing line in the event
of a transmission error.

Therefore when a transmission error occurs due to a bad communication link, the group 3 I D can be
used to synchronize and correct the error.

Data formatting for CClIT Group 3 2D

The 2D scheme uses a combination of additional codes called vertical code, pass code, and horizontal
code to encode every line in the group of k lines.

The steps for pseudocode to code the code line are:

(i) Parse the coding line and look for the change in the pixel value. (Change is found at al location).

(ii) Parse the reference line and look for the change in the pixel value. (Change is found at bl location).

(iii) . Find the difference in location between bland a 1: delta = b1- al

Advantage of CClIT Group 3 2D

The implementation of the k factor allows error-free

transmission . Compression ratio achieved is better than CClTT Group 3 1 D . It is accepted for document
imaging applications.

DisadvantageIt doesn’t provide dense compression

CCITT Group 4 2D compression

CClTT Group 4 compression is the two dimensional coding scheme without the k-factor.

In this method, the first reference line is an imaginary all-white line above the top of the image.The first
group of pixels (scanline) is encoded utilizing the imaginary white line as the reference line.

The new coded line becomes the references line for the next scan line. The k-factor in this case is the
entire page of line. In this method, there are no end-of-line (EOL) markers before the start of the
compressed data

COLOR, GRAY SCALE AND STILL-VIDEO IMAGE COMPRESSION

Color:

Color is a part of life we take for granted. Color adds another dimension to objects. It helps in making
things standout.

Color adds depth to images, enhance images, and helps set objects apart from -background.

Let us review the physics of color. Visible light is a form of electromagnetic radiation or radiant energy,
as are radio frequencies or x-rays. The radiant energy spectrum contains audio frequencies, radio
frequencies, infrared, visible light, ultraviolet rays, x-rays and gamma rays.

Radian energy is measured in terms of frequency or wavelength. The relationship between the two is
Color Characteristics

We typically define color by its brightness, the hue and depth of color.

Luminance or Brightness

This is the measure of the brightness of the light emitted or reflected by an object; it depends on the
radiant, energy of the color band.

Hue This is the color sensation produced in an observer due to the presence of certain wavelengths
of color. Each wavelength represents a different hue.

Saturation This is a measure of color intensity, for example, the difference between red and pink. Color
Models Several calm' models have been developed to represent color mathematically. Chromacity
Model It is a three-dimensional model with two dimensions, x and y, defining the color, and the third
dimension defining the luminance. It is an additive model since x and yare added to generate different
colors.

RGBModel RGB means Red Green Blue. This model implements additive theory in that
different intensities of red, green and blue are added to generate various colors.
HSI Model The Hue Saturation and Intensity (HSI) model represents an artist's impression of tint,
shade and tone. This model has proved suitable for image processing for filtering and smoothing
images. CMYK Model The Cyan, Magenta, Yellow and Black color model is used in desktop
publishing printing devices. It is a color-subtractive model and is best used in color printing devices only.

YUV Representation The NTSC developed the YUV three-dimensional color model. y -Luminance
Component

UV -Chrominance Components.

Luminance component contains the black and white or grayscale information. The chrominance
component contains color information where U is red minus cyan and V is megenta minus green.

YUV Model for JPEG

The JPEG compression scheme uses several stages.

The first stage converts the signal from the spatial RGB domain to the YUV frequency domain by
performing discrete cosine transform. This process allows separating luminance or gray-scale
components from the chrominance components of the image.

JOINT PHOTOGRAPHIC EXPERTS GROUP COMPRES5ION (JPEG)

ISO and CCITT working committee joint together and formed Joint Photographic Experts Group. It is
focused exclusively on still image compression.

Another joint committee, known as the Motion Picture Experts Group (MPEG), is concerned with full
motion video standards.

JPEG is a compression standard for still color images and grayscale images, otherwise known as
continuous tone images.
JPEG has been released as an ISO standard in two parts

Part I specifies the modes of operation, the interchange formats, and the encoder/decoder specifies for
these modes along with substantial implementation guide lines.

Part 2 describes compliance tests which determine whether the implementation of an encoder or
decoder conforms to the standard specification of part I to ensure interoperability of systems compliant
with JPEG standards

Requirements addressed by JPEG

The design should address image quality.

The compression standard should be applicable to practically any kind of continuous-tone digital source
image .

It should be scalable from completefy lossless to lossy ranges to adapt it. It should provide sequential
encoding .

It should provide for progressive encoding .

It should also provide for hierarchical encoding .

The compression standard should provide the option of lossless encoding so that images can be
guaranteed to provide full detail at the selected resolution when decompressed.

Definitions in the JPEG Standard

The JPEG Standards have three levels of definition as follows:

* Base line system

* Extended system

* Special lossless function.

The base line system must reasonably decompress color images, maintain a high compression ratio, and
handle from 4 bits/pixel to 16 bits/pixel.

The extended system covers the various encoding aspects such as variable-length encoding, progressive
encoding, and the hierarchical mode of encoding.

The special lossless function is also known as predictive lossless coding. It ensures that at the resolution
at which the image is no loss of any detail that was there in the original source image.

Overview of JPEG Components JPEG Standard components are:

(i) Baseline Sequential Codec

(ii) OCT Progressive Mode

(Hi) Predictive Lossless Encoding (iv) Hierarchical Mode.

These four components describe four four different levels of jPEG compression.

The baseline sequential code defines a rich compression scheme the other three modes describe
enhancements to this baseline scheme for achieving different results.

Some of the terms used in JPEG methodologies are:

Discrete Cosine Transform (OCT)

OCT is closely related to Fourier transforms. Fourier transforms are used to represent a two dimensional
sound signal.

DCT uses a similar concept to reduce the gray-scale level or color signal amplitudes to equations that
require very few points to locate the amplitude in Y-axis X-axis is for locating frequency.
DCT Coefficients

The output amplitudes of the set of 64 orthogonal basis signals are called OCT Co-
efficients. Quantization This is a process that attempts to determine what information can be safely
discarded without a significant loss in visual fidelity. It uses OCT co-efficient and provides many-to-one
mapping. The quantization process is fundamentally lossy due to its many-to-one mapping.

De Quantization This process is the reverse of quantization. Note that since quantization used a many-
to-one mapping, the information lost in that mapping cannot be fully recovered

Entropy Encoder / Decoder Entropy is defined as a measure of randomness, disorder, or chaos, as well
as a measure of a system's ability to undergo spontaneous change. The entropy encoder compresses
quantized DCT co-efficients more compactly based on their spatial characteristics. The baseline
sequential. codec uses Huffman coding. Arithmetic coding is another type of entropy
encoding Huffman Coding Huffman coding requires that one or more sets of huff man code tables be
specified by the application for encoding as well as decoding. The Huffman tables may be pre-defined
and used within an application as defaults, or computed specifically for a given image.

Baseline Sequential codec

It consists of three steps: Formation of DCT co-efficients quantization, and entropy encoding. It is a rich
compression scheme.

DCT Progressive Mode

The key steps of formation of DCT co-efficients and quantization are the same as for the baseline
sequential codec. The key difference is that each image component is coded in multiple scans instead
ofa single scan.
Predictive Lossless Encoding

It is to define a means of approaching lossless continuous-tone compression. A predictor combines

sample areas and predicts neighboring areas on the basis of the sample areas. The predicted areas are
checked against the fully loss less sample for each area.

The difference is encoded losslessly using huffman on arithmetic entropy encoding .

Hierarchical Mode

The hierarchical mode provides a means of carrying multiple resolutions. Each successive encoding of
the image is reduced by a factor of two, in either the horizontal or vertical dimension.

JPEG Methodology

The JPEG compression scheme is lossy, and utilizes forward discrete cosine transform (or forward DCT
mathematical function), a uniform quantizer, and entropy encoding. The DCT function removes data
redundancy by transforming data from a spatial domain to a frequency domain; the quantizer quantizes
DCT co-efficients with weighting functions to generate quantized DCT co-efficients optimized for the
human eye; and the entropy encoder minimizes the entropy of quantized DCT co-efficients.

The JPEG method is a symmetric algorithm. Here, decompression is the exact reverse process of
compression.

Figure below describes a typical DCT based encoder and decoder. Symmetric Operation of DCT based
Codec
Figure below shows the components and sequence of quantization 5 * 8

Image blocks

Quantization

Quantization is a process of reducing the precision of an integer, thereby reducing the number of bits
required to store the integer, thereby reducing the number of bits required to store the integer.

The baseline JPEG algorithm supports four color quantization tables and two huffman tables for both DC
and AC DCT co-efficients. The quantized co-efficient is described by the following equation:

ZigZag Sequence

Run-length encoding generates a code to represent the C0unt of zero-value OCT co-efficients. This
process of run-length encoding gives an excellent compression of the block consisting mostly of zero
values.
Further empirical work proved that the length of zero values in a run can be increased to give a further
increase in compression by reordering the runs. JPEG came up with ordering the quantized OCT co-
efficients in a ZigZag sequence

Entropy Encoding

Entropy is a term used in thermodynamics for the study of heat and work. Entropy, as used in data
compression, is the measure of the information content of a message in number of bits. It is
represented as

Entropy in number of bits = log2 (probability of Object)

VIDEO IMAGE COMPRESSION

The development of digital video technology has made it possible to use digital video compression for a
variety of telecommunications applications. Standardization of compression algorithms for video was
first initiated by CCITT for teleconferencing and video telephony

Multimedia standards for Video:

Requirements for full-motion Video Compression

Applications using MPEG standards can be symmetric or asymmetric. Symmetric applications are
applications that require essentially equal use of compression and decompression. Asymmetric
applications require frequent decompression.

Symmetric applications require on-line input devices such as video cameras, scanners and microphones
for digitized sound. In addition to video and audio compression, this standards activity is concerned with
a number of other Issues concerned with playback of video clips and sound clips. The MPEG standard
has identified a number of such issues that have been addressed by the standards activity. Let us review
these Issues.

Random Access

The expectations generated for multimedia systems are the ability to playa sound or video clip from any
frame with that clip, irrespective of on what kind-of media the information is stored

VCR paradigm

The VCR paradigm consists of the control functions typically found on a VCR such as play, fast forward,
rewind, search forward and rewind search.

Multiplexing Multiple Compressed Audio and Video Bit Streams

It is a special requirement retrieved from different storage centers on a network. It may be received
from different storage centers on a network. It may have to be achieved in a smooth manner to avoid
the appearance of a jumpy screen.

Editability

Playback Device Flexibility

CCITT H.261 Video Coding Algorithms (P x 64)

The linear quantizer uses a step algorithm that can be adjusted based on picture quality and coding
efficiency. The H.261 is a standard that uses a hybrid of OCT and OPCM (differential pulse Code
Modulation) schemes with motion estimation.

It also defines the data format. Each MB contains the OCT coefficients (TCOEFF) ofa block followed by an
EOB (a fixed length end-of-block marker). Each MB consists of block data and an MB header. A GOB
(Group of Blocks) consists of a GOB header. The picture layer consists of a picture header. The H.261 is
designed for dynamic use and provides a fully contained organization and a high level of interactive
control.

Moving Picture Experts Group Compression

The MPEG standards consist of a number of different standards.

The MPEG 2 suite of standards consist of standards for MPEG2 Video, MPEG - 2 Audio and MPEG - 2
systems. It is also defined at different levels, called profiles.

The main profile is designed to cover the largest number of applications. It supports digital video
compression in the range of2 to 15 M bits/sec. It also provides a generic solution for television
worldwide, including cable, direct broadcast satellite, fibre optic media, and optical storage media
(including digital VCRs).

MPEG Coding Methodology

The above said requirements can be achieved only by incremental coding of successive frames. It is
known as interframe coding. If we access information randomly by frame requires coding confined to a
specific frame, then it is known as intraframe coding.
The MPEG standard addresses these two requirements by providing a balance between interframe
coding and intraframe coding. The MPEG standard also provides for recursive and non-recursive
temporal redundancy reduction.

The MPEG video compression standard provides two basic schemes: discrete-transform-based
compression for the reduction of' spatial redundancy and block-based motion compensation for the
reduction of temporal (motion) redundancy. During the initial stages of DCT compression, both the full
motion MPEG and still image JPEG algorithms are essentially identical. First an image is converted to the
YUVcolor space (a luminance/chrominance color space similar to that used for color television). The
pixel data is then fed into a discrete cosine transform, which creates a scalar quantization (a two-
dimensional array representing various frequency ranges represented in the image) of the pixel data.

Following quantization, a number of compression algorithms are applied, including run-length and
Huffman encoding. For full motion video (MPEG I and 2), several more levels of block based motion-
compensated techniques are applied to reduce temporal redundancy with both causal and noncausal
coding to further reduce spatial redundancy.

The MPEG algorithm for spatial reduction is lossy and is defined as a hybrid which employs motion
compensation, forward discrete cosine transform (DCF), a uniform quantizer, and Huffman coding.
Block-based motion compensation is utilized for reducing temporal redundancy (i.e. to reduce the
amount of data needed to represent each picture in a video sequence). Motion-compensated reduction
is a key feature of MPEG.

Moving Picture Types

Moving pictures consist of sequences of video pictures or t1'ame'S that are played back a fixed number
of frames per second. To achieve the requirement of random access, a set of pictures can be defined to
form a group of pictures (GOP) consisting of one or more of the following three types of pictures.

Intra pictures (1)

Undirectionally predicted pictures (U) Bidirectiomflly predicted pictures (B)

A Gap consists of consecutive pictures that begin with an intrapicture. The intrapicture is coded without
any reference to any other picture in the group.

PrediGted pictures are coded with a reference to a past picture, either an intrapicture or a
unidirectionally predicted picture. Bidirectionally predicted picture is never used as referencesMotion
Compensation for Coding MPEG

Let us review the concept of Macroblocks and understand the role they play in compression

MACRO BLOCKS

For the video coding algorithm recommended by CCITT, CIF and QCIF are diviqed into a hierarchical
block structure consisting of pictures, groups of blocks (GOBs), Macro Blocks(MBs), and blocks. Each
picture frame is divided into 16 x 16 blocks. Each Macroblock is composed of four 8 x 8 (Y) luminance
blocks and two 8 x 8 (Cb and Cn) chrominance blocks. This set of six blocks, called a macroblock; is the
basic hierarchical component used for achieved a high level of compression.

Motion compensation

Motion compensation is the basis for most compression algorithms for visual telephony and full-motion
video. Motion compensation assumes that the current picture is some translation of a previous picture.
This creates the opportunity for using prediction and interpolation. Prediction requires only the current
frame and the reference frame.

Based on motion vectors values generated, the prediction approach attempts to find the relative new
position of the object and confirms it by comparing some block exhaustively. In the interpolation
approach, the motion vectors are generated in relation to two reference frames, one from the past and
the next predicted frame.

The best-matching blocks in both reference frames are searched, and the average is taken as the
position of the block in the current frame. The motion vectors for the two reference, frames are
averaged.

Picture Coding Method

In this coding method, motion compensation is applied bidirectionally. In MPEG terminology, the
motion-compensated units are called macro blocks (MBs).

MBs are 16 x 16 blocks that contain a number of 8 x 8 luminance and chrominance blocks. Each 16 x 16
macro block can be of type intrapicture, forward-predicted, backward predicted, or average.

MPEG Encoder

Figure below shows the architecture of an MPEG encoder. It contains DCT quantizer, Huffman coder and
Motion compensation. These represent the key modules in the encoder.

Architecture of MPEG Encoder:

The Sequence of events for MPEG

First an image is converted to the YUV color space.

The pixel data is then fed into a DCT, which creates a scalar quantization of the pixel data.

Following quantization, a number of compression algorithms are applied, including run-length and
Huffman encoding. For full-motion video, several more levels of motion compensation compression and
coding are applied.

MPEG -2

It is defined to include current television broadcasting compression and decompression needs, and
attempts to include hooks for HDTV broadcasting.

The MPEG-2 Standard Supports:

1.Video Coding: * MPEG-2 profiles and levels.

2.Audio Coding:*MPEG-l audio standard fro backward compatibility.

* Layer-2 audio definitions for MPEG-2 and stereo sound.

* Multichannel sound.

3. Multiplexing: MPEG-2 definitions

MPEG-2, "The Grand Alliance"

It consists of following companies AT&T, MIT, Philips, Sarnoff Labs, GI Thomson, and Zenith.

The MPEG-2committee and FCC formed this alliance. These companies together have defined the
advanced digital television system that include the US and European HDTV systems. The outline of the
advanced digital television system is as follows:

1.Format: 1080/2: 1160 or 720/1.1160

2.Video coding: MPEG-2 main profile and high level 3.Audio coding: Dolby AC3

4.Multiplexor: As defined in MPEG-2

Modulation: 8- VSB for terrestrial and 64-QAM for cable.

Vector Quantization

Vector quantization provides a multidimensional representation of information stored in look-up tables,

vector quantization is an efficient pattern-matching algorithm in which an image is decomposed into two
or more vectors, each representing particular features of the image that are matched to a code book of
vectors.

These are coded to indicate the best fit.

In image compression, source samples such as pixels are blocked into vectors so that each vector
describes a small segment or sub block of the original image.

The image is then encoded by quantizing each vector separately

Intel's Indeo Technology

It is developed by Intel Architecture Labs Indeo Video is a software technology that reduces the size of
uncompressed digital video files from five to ten times.

Indeo technology uses multiple types of 'lossy' and 'lossless' compression techniques.

AUDIO COMPRESSION

Audio consists of analog signals of varying frequencies. The audio signals are converted to digital form
and then processed, stored and transmitted. Schemes such as linear predictive coding and adaptive
differential. pulse code modulation (ADPCM) are utilized for compression to achieve 40-80%
compression.

FRACTAL COMPRESSION

A fractal is a multi-dimensional object with an irregular shape or body that has approximately the same
shape or body irrespective of size. For example, if you consider 'stick' as your object, the fractal is
defined, mathematically as

where L - approaches 0,

N(L) ~ number of stick L, and L is the length of the stick.