Unit 7

1) What is
Image compression?
Ans)
Image compression is the
process of
reducing the amount of data required
to
represent an
image.

Image Compression
Transmit (channes
fre,y) -> Compress > Decompress fic,y)
-

2 f
store
retrieve

storage
device

compression provides
Image and case of storage
transmission.

Uncompressed original data

BBBBAAA2I2

compressed data
#
B04A03204
Image
10 10

20
20
Compressed data 10032006.
-

20 20

1 20 10

A
Compression Fundamentals

Image training
compression reduce of
s the nation
ize
·

Retaining info depends on the applications

compression: uncompressed image fill
ratio
compressed image file.

compression can be achieved as we

have a lot redundant data

of in
any
information
·

Dater-information + redundant data.

⑪ Data
redundancy
-
no relevant information
-

repetitive data.

Compression -> reduce amount

of data
by
reducing the
redundancy.

featuresshouldbe unchange of
· the

Types or

redundancy L
Spatial Temporal
redundancy redundancy
① Spatial Redundancy (for image & video)

videos
-
Present in
of images and
areas

pixel values
values
by small thin
·

vary are

a structure

Examples - of blue
Image sky
-

⑧ Temporal Redundancy (for video

Present video where there
-
in
signals,
is
similarity between two frames
-

If frames are similar and small, it

is
inefficient to
compress a video
series
signal as a
of image.

* Basic Data Redundancies

Coding redundancy
redundan
·.
Psycho visual
redundancy.
·
*
Coding redundancy
·
Information represented in different
codes.
forms of
·

Coding techniques to convert

information
into acode.

code
Technique assigns unique
·
a

to
symbols of message/pixels of image.

wrongchoicecreates
addictionis
unnecessary
·

called
extra bite are
redundancy.
·

⑤Inter pixel
redundancy.
·

Neighbouring pinels in the same

image
are correlated

· Inter pinel correlation is called as

inter-pixel spatial redundancy.

· Inter pixel redundancy in successive
frames of videointer pinel temporal
a

redundary
· single pine is redundant, can be guessed from
 the values of its
neighbours,

Psycho-visual redundancy
A

·
related to vision

eliminate levels that do not

grey
·

affect the interpretation of the

image,
that
Deals
information
with is
·

redundant based, on human vision

levels reduced by
Eg: grouping to
· 256 are
16 levels, objects are
still
recognizable.
This irreversible.
compression is
·

A
Fidelity
·
The
degree of cractness with which
the data is reproduced.
used to determine important information
·

retain when
to
compressing file
·

Fidelity
1
Criteria.
objective fidelity criteria
criteria.
2 subjective fidelity
Objective
·
fidelity Criteria.

Measure the amount inthe

1.
of error

decompressed image.
2. Smaller the error value, better
the
quality of the compressed image.

Error Metrics / Objective

Fidelity.Criteria
·

Mean squared error.

to ratio
signal noise
-

ratio.
-
Peak
signal to noise
* Mean squared error.
used data.
for 24
->

fluzy) and reconstructed

image - "luy)
-

emse: () Eifencys-fansys]
M: number ofrows

N= number of columns
·van
squared error

come:
E Ettines cases)
-

m= number
of news
column
n number
of
=

Signal to Noise Ratio (SNR)

SNR.: Signal power: Is
Noise power Pa

input image, fu, y)

signal power-m* ifrmp)

Reconstructed f(n, y)
=

image
 esrz=5 I
JEEERis Peace
E"z [flecy)-fccessa
· Peak
Signal to Noise Ratio

PSNR=10logi E5
·

Subjective Fidelity Criteria

Excellent
&
-high quality of image

3
Fine
Image of acceptable
2.

qualityinterfere
Passable
waitasonable
3.

4.
Marginal
Inferior imagequalityobjectional
is
5.
poor
-

6. Unusable- Bad viewed.

image, cannot be
 2x2 Find RMSC.
image.
*

Original Reconstructed
54 4 4
2 B I 2

2
RMSE=V1/4 (5-4)2 +14-4)2 + 12-117+ 13-2)

=Fir
General Image Compression Modele
·

redundan
encoderremove test of symbols
Channel encoder improves the noise
immunity
-
·

robustness
ensure
against
-

channel noise.
 Channel decoder - back the
original
·

decoder
get
source
signal

Run
·

or
length
reduce
is used to
eliminating
coding interpixel redundancy.
is used to eliminate
Huffman encoding
·

and reduce redundancies

coding

③General Image Compression Modele

·

Compression types: lossy, Lossless

consists
image compression system of 2 main
·

blocks

encoder compressor
anout
·

image f(y) -> i encoded by the encoder

making it easy for transmission.

·
The decoder receives the transmitted
and
signal reconstructs the output
image fluzy).
 ·

If the system is an error

free one

fee,y) will be a replica of fluzy)

A
Classification of data Compression techniques
X Reconstruction
#
>Compression XC
- >
Algorithm Algorithm
Compression

Error free
L
sy
compression
Compression
x FY
X =
Y

·
Lossless Data Compression
-

Integrity of the data is preserved.

Data after compression and
-

decompression are the same

is lost.
-
No part of data
·

Lossy Data Compression

-
Loss of some
fidelity is acceptable,

critical data.by losing

-
Data is
compressed or

some

to
aims the
minimize
date
a
-
·
Lossless Compression Techniques
1.
Dictionary
based techniques
2. Statistical based techniques

Based Technique
1.
Dictionary
·

Images consist of recurring patterns

is to keep
·
efficient way dictionary a

of frequently occuring patterns. -

@ when these
encoded
patterns
with
appear
to
they are a
reference
the
dictionary
We split the
image pixels into two
·

parts
frequently occuring patterns
->
infrequently occuring patterns
-

Example of Dictionary based encoding.

Run length
coding
·

2. B 5 6
6,4
H I -
H

·
first code
specifies the
gray
value
followed by length of run.

Hence the RLE of the

first row -235664

* Drawbacks of RLC

-
At times, double of size of file.
Example:1, 2, 5, 3,
·
12.

1121513112 I
·

RLE code is double than that of original

sequence
 ·

Hence, RLE should only be used if

we have the same
grey
value repeated.

Statistical method of
lossless
compression technique

·Huffman coding
· It is a lossless statistical method
· most
often
bits,
characters are
assigned
few
rarely characters are
occuring
·

less bits
assigned
·

steps of Huffman coding.

① for the
given grey levels, write down the
list of probabilities
of sort the list in
descending order

⑧ Add the smallest (last) I probabilities

& the
probabilities and
Again the
sort

repeat
value is left.
process until just one
 a) Compute
date
Huffman code
for the
given
·

E1, 1, 1, 1, 1, 3, 1,2,2,2,2,2,2, 3, 3, 3, 3, 4, 4, 4. 4,

5, 5, 5, 6, 6, 73
S; M; Pi=n;/ total count

I
T 7/28 = 0.25

3
2
6/28 = 0.21

(28 0.18

3
5 =

4 128 =
0.14
3/28 =
0.11

2
2/28
=
0.07

eB
7 I 1/28 =
0.04

So Pi D

I 0.25
z
2 0.21

G 0.18
4 0.14
5 8 II

G ooz0.22 -
Do 1I
#7 80 C
 I
I ⑦

0.57 0.43
⑧
I 8
I

0.32 0.25 002l

0.22
o
S D
1
/
8.18 0014
Do I
Poll
O
I
0.07 0.04

Si Huffman Code

I 1 O

2 &O

3
1 I I
110
01 I

G O10)
O 10 O
 Image Compression standards.

Data Compression
methods

lossless Methods lossy methods

. Y
Run Huffman Tempel
length ziv JPtG MPtG Mp3

Lossy compression
Many
·
methods of
lossy compression have
been developed
· A
family ofhas
techniques known Transform
as

to be most valuable.
coding proven

Transform coding method

·

·
Discard the
pinel coefficients that are

nearly zero.
 reconstructed
When
images are
using
·

inverse transform, we get compressed

image in which little transformation
is lost.

·Image compression standards

continuous tone
images.
·

JPEG -

compress and code continues tone

-

and
monochecome color images of res

and to standard
-very simple
based easy discrete
use

that is on cosine

transform
- does not handle bi-level (black &
White
images very well.
Advantage of JPEG, allowing the
-

user

2 main modes: called baseline

Cossy:
-

lossless: produces compression

ration
 -Most
mode
implementations support just the
lossy
mode includes &
lossy progressive
-

hierarchial coding.

JPEG, compression method, not a

-

complete standard for image

representation.

Image Compression for Binary image

-

Binary image only black & white pixels

-

Also called bi-level.

->

each
pinel is stored as a
single bit
be stored
binary image can in
-

memory
as a bitmap.
-

Binary
well
images
with
also
compress
simple run
length.
Binary
image compress well with
-

simple run length compression schemes