45
Digital Image Processing, 2nd ed. Digital Image Processing, 2nd ed. www.imageprocessingbook.com © 2002 R. C. Gonzalez & R. E. Woods Image Compression Outline – Fundamentals Redundancies Image compression models Information theory impact Entropy Coding theorems Loss-less compression Variable-length coding Bit-place coding Predictive coding – Lossy compression Predictive coding Transform coding Compression standards

No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

  • Upload
    others

  • View
    10

  • Download
    0

Embed Size (px)

Citation preview

Page 1: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Imag

e C

ompr

essi

on

•O

utlin

e–

Fund

amen

tals

•R

edun

danc

ies

–Im

age

com

pres

sion

mod

els

–In

form

atio

n th

eory

impa

ct•

Entro

py•

Cod

ing

theo

rem

s–

Loss

-less

com

pres

sion

•V

aria

ble-

leng

th c

odin

g•

Bit-

plac

e co

ding

•Pr

edic

tive

codi

ng–

Loss

yco

mpr

essi

on•

Pred

ictiv

e co

ding

•Tr

ansf

orm

cod

ing

–C

ompr

essi

on st

anda

rds

Page 2: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Fundam

enta

lsFu

ndam

enta

ls

•Dat

a co

mpr

essi

on: p

roce

ss o

f red

ucin

g th

e am

ount

of d

ata

requ

ired

to re

pres

ent a

giv

en q

uant

ity o

f in

form

atio

n.

•Dat

a an

d in

form

atio

n

•Dat

a co

nvey

s inf

orm

atio

n

•Dat

a re

dund

ancy

•Giv

en tw

o da

ta se

ts w

ith n

umbe

rs o

f inf

orm

atio

n-ca

rryi

ng u

nits

n1

and

n 2, c

orre

spon

ding

ly.

•Com

pres

sion

ratio

: CR=

n 1/n

2

•Rel

ativ

e da

ta re

dund

ancy

: RD=1

–1/

CR

•Thr

ee ty

pes o

f dat

a re

dund

ancy

•Cod

ing

redu

ndan

cy

•Int

erpi

xelr

edun

dand

ancy

•Psy

chov

isua

lred

unda

ncy

Page 3: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Fundam

enta

ls:

Codin

g R

edundan

cyFu

ndam

enta

ls:

Codin

g R

edundan

cy

•His

togr

am a

naly

sis o

f gre

y-sc

ale

imag

es

•Giv

en d

iscr

ete

rand

om v

aria

ble

r k∈

[0,1

] rep

rese

ntin

g L

grey

leve

ls; k

=0,1

,2,…

L–

1;

each

r koc

curs

with

pro

babi

lity

p r(r

k), i.

e. p

r(rk)=

nk

/n, w

here

nk

is th

e nu

mbe

r of t

imes

th

at k

-thgr

ey le

vel a

ppea

rs a

nd n

is th

e to

tal n

umbe

r of p

ixel

s.

•Giv

en l(

r k) is

the

num

ber o

f bits

to re

pres

ent e

ach

valu

e of

r k. A

vera

ge n

umbe

r of b

its

requ

ired

to re

pres

ent e

ach

pixe

l is

•Tot

al n

umbe

r of b

its to

cod

e an

Nim

age

is M

NL av

g

•Exa

mpl

e: 8

-leve

l im

age

code

d by

3-b

it bi

nary

cod

e an

d a

varia

ble-

leng

th c

ode

∑= =

=1 0

)(

)(

L kk

rk

avg

rp

rl

L

099

.011.1/1

111.1

7.2/3

7.2)

()

(

3)

(3

7

02

7

01

=−

==

=

==

==

∑∑==

DR

kk

rk

kk

r

RC

rp

rl

L

rp

L

avg

avg

Page 4: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Fundam

enta

ls:

Codin

g R

edundan

cyFu

ndam

enta

ls:

Codin

g R

edundan

cy

•The

fund

amen

tal b

asis

of d

ata

com

pres

sion

thro

ugh

vari

able

-leng

th c

odin

g: fu

nctio

ns

p r(r

k) an

d l(

r k) ar

e in

vers

ely

prop

ortio

nal.

•Cod

e re

dund

ancy

occu

rs w

hen

the

code

doe

s not

min

imis

e th

e av

erag

e le

ngth

(not

full

adva

ntag

e of

pro

babi

litie

s is t

aken

)

Page 5: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Fundam

enta

ls:

Inte

rpix

elRed

undan

cyFu

ndam

enta

ls:

Inte

rpix

elRed

undan

cy

•Rel

ated

to th

e in

terp

ixel

corr

elat

ions

w

ithin

an

imag

e

•spa

tial r

edun

danc

y, g

eom

etric

re

dund

ancy

, int

erfr

ame

redu

ndan

cy

•Mea

sure

d th

roug

h au

toco

rrel

atio

n co

effic

ient

s

•Tra

nsfo

rmat

ions

that

rem

ove

inte

rpix

elre

dund

ancy

are

cal

led

map

ping

s

∑∆

−−

=

∆+

∆−

=∆

∆=

nN

yn

yx

fy

xf

nN

nAw

here

An

An

1

0)

,(

),

(1

)(

)0(/)

()

Page 6: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Fundam

enta

ls:

Inte

rpix

elRed

undan

cyFu

ndam

enta

ls:

Inte

rpix

elRed

undan

cy

Exam

ple

of a

non

-vis

ual m

appi

ng:

•102

4x34

3 gr

ey-s

cale

imag

e tra

nsfo

rmed

to a

bi

nary

imag

e

•Eac

h lin

e is

repr

esen

ted

by a

sequ

ence

of p

airs

(g

i, w

i), w

here

gii

s the

i-th

leve

l and

wii

s the

nu

mbe

r of s

ucce

ssiv

e sa

mpl

e of

this

leve

l (ru

n le

ngth

).

•For

line

100

onl

y 88

bits

are

nee

ded

to re

pres

ent

the

1024

bits

of b

inar

y da

ta

•For

the

who

le im

age

1216

6 ru

ns

are

suff

icie

nt

•11

bits

repr

esen

t eac

h ru

n-le

ngth

pa

ir

62.063.2

11

63.211.

1216

61.

343

.10

24

=−

=

==

DR

RC

Page 7: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Fundam

enta

ls:

Psy

cho-v

isual

Red

undan

cyFu

ndam

enta

ls:

Psy

cho-v

isual

Red

undan

cy

•Eye

doe

s not

resp

ond

with

equ

al se

nsiti

vity

to a

ll vi

sual

info

rmat

ion.

Cer

tain

info

rmat

ion

sim

ply

has l

ess r

elat

ive

impo

rtanc

e. T

his i

nfor

mat

ion

is sa

id to

be p

sych

o-vi

sual

ly

redu

ndan

t. It

can

be

elim

inat

ed w

ithou

t sig

nific

antly

impa

ring

the

qual

ity o

f im

age

perc

eptio

n.

•Som

e kn

owle

dge

abou

t how

the

brai

n re

cogn

izes

pic

ture

s is n

eede

d

•Psy

cho-

visu

al re

dund

ancy

diff

ers f

rom

the

prev

ious

two.

It is

ass

ocia

ted

with

real

vis

ual

info

rmat

ion.

Its e

limin

atio

n re

sults

in a

loss

of q

uant

itativ

e in

form

atio

n an

d it

is re

ferr

ed to

as

qua

ntiz

atio

n. It

is a

n irr

ever

sibl

e op

erat

ion

and

lead

s to

loss

yda

ta c

ompr

essi

on.

•Exa

mpl

e: im

prov

ed g

rey-

scal

e (I

GS)

qua

ntiz

atio

n.

Page 8: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Fundam

enta

ls:

Psy

cho-v

isual

Red

undan

cyFu

ndam

enta

ls:

Psy

cho-v

isual

Red

undan

cy

Page 9: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Fundam

enta

ls:

Fidel

ity

Crite

ria

Fundam

enta

ls:

Fidel

ity

Crite

ria

•Obj

ectiv

ean

d su

bjec

tive

fidel

ity c

riter

ia to

ass

ess t

he in

form

atio

n lo

ss

•Roo

t-mea

n-sq

uare

err

or

or

M

ean-

squa

re si

gnal

-to-n

oise

ratio

• •Rat

ing

scal

es(s

ee T

able

8.3

) or

Sid

e-by

-sid

e co

mpa

rison

s

{-3,

-2, -

1, 0

, 1, 2

, 3}≡

{muc

h w

orse

, wor

se, s

light

ly w

orse

, the

sam

e, sl

ight

ly b

ette

r, be

tter,

muc

h be

tter}

∑∑∑

∑∑

∑− =

− =− =

− =− =

− =−

=

=1 0

1 02

1 0

1 02

2/11 0

1 02

)],

(ˆ)

,(

[

)],

([

)],

(ˆ)

,(

[1

M x

N y

M x

N ym

sM x

N yrm

sy

xf

yx

f

yx

fSN

Ry

xf

yx

fM

Ne

Page 10: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Imag

e Com

pre

ssio

n M

odel

sIm

age

Com

pre

ssio

n M

odel

s

•A g

ener

al c

ompr

essi

on sy

stem

con

sist

s of a

n en

code

r and

a d

ecod

er.

•Sou

rce

enco

der (

and

deco

der)

: res

pons

ible

for r

educ

ing

or e

limin

atin

g re

dund

anci

es.

•Cha

nnel

enc

oder

(and

dec

oder

): re

duce

s the

impa

ct o

f cha

nnel

noi

se b

y ‘c

ontro

lled

redu

ndan

cy’.

Exam

ple:

Ham

min

g(7

,4) c

ode

Inter

pixel

rdnd

cy→

Psyc

ho-v

isual

rdnd

cy→

←co

ding

rdnd

cy

16

01

24

25

01

32

33

02

31

01

23

nu

mbe

r bi

t -4 a

give

n

bh

bb

bh

bh

bb

bh

bh

bb

bh

bb

bb

=⊕

⊕=

=⊕

⊕=

=⊕

⊕=

76

53

76

54

1

76

33

2

75

31

1

12

4

w

ord

code

co

rrec

ted

th

e

w

ord

nonz

ero

aby

ch

eck

parit

y

hh

hh

hh

hh

ch

hh

hc

hh

hh

cc

cc

⊕⊕

⊕=

⊕⊕

⊕=

⊕⊕

⊕=

Page 11: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Ele

men

ts o

f In

form

atio

n T

heo

ryEle

men

ts o

f In

form

atio

n T

heo

ry

Mea

surin

g in

form

atio

n

The

gene

ratio

n of

info

rmat

ion

is m

odel

ed a

s a p

roba

bilis

tic p

roce

ss. R

ando

m e

vent

Eoc

curs

with

pro

babi

lity

P(E)

The

base

of t

he lo

garit

hm d

eter

min

es th

e un

its u

sed

to m

easu

re th

e in

form

atio

n. If

the

base

2

is se

lect

ed th

e re

sulti

ng in

form

atio

n un

it is

cal

led

bit.

If P

(E)=

0.5

(two

poss

ible

equ

ally

lik

ely

even

ts) t

he in

form

atio

n is

one

bit.

)(

log

)(1

log

)(

EP

EP

EI

−=

=

Page 12: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Ele

men

ts o

f In

form

atio

n T

heo

ryEle

men

ts o

f In

form

atio

n T

heo

ry

•The

info

rmat

ion

chan

nel:

the

phys

ical

med

ium

that

link

s the

sour

ce to

the

user

(tel

. lin

e,

wire

, ele

ctro

mag

netic

wav

es, e

tc.)

•The

sour

ce•M

odel

ed a

s a d

iscr

ete

rand

om v

aria

ble

•Sou

rce

alph

abet

A={

a j}

•Sym

bols

(let

ters

) ajw

ith p

roba

bilit

iesP

(aj)

•The

ave

rage

self-

info

rmat

ion

obta

ined

from

kou

tput

s is

•The

ave

rage

info

rmat

ion

per s

ourc

e ou

tput

is

•(un

cert

aint

yor

ent

ropy

)

∑=

−=

=−

−−

−J j

jj

JJ

aP

aP

k

aP

akP

aP

akP

aP

akP

1

22

11

)(

log

)(

)(

log

)(

...)

(lo

g)

()

(lo

g)

(

∑=

−=

J jj

ja

Pa

PH

1)

(lo

g)

()

(z

Page 13: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Ele

men

ts o

f In

form

atio

n T

heo

ryEle

men

ts o

f In

form

atio

n T

heo

ry

•The

cha

nnel

•The

info

rmat

ion

in th

e ou

tput

of t

he c

hann

el is

a d

iscr

ete

rand

om v

aria

ble

•Cha

nnel

alp

habe

t B={

b k}

with

lette

rs b

kw

ith p

roba

bilit

iesP

(bk)

•The

pro

babi

litie

s are

rela

ted

to th

e so

urce

z b

y co

nditi

onal

pro

babi

litie

s

•Cha

nnel

(tra

nsiti

on) m

atrix

==

J jj

jk

ka

Pa

bP

bP

1)

(lo

g)

|(

)( [

])

|(

w

here

,)

|(

...)

|(

)|

(......

)|

()

|(

...)

|(

)|

(

211

2

12

11

1

jk

kjkj

JK

KK

J

ab

Pq

qa

bP

ab

Pa

bP

ab

Pa

bP

ab

Pa

bP

==

⋅⋅

⋅⋅

⋅=

QQ

Page 14: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Ele

men

ts o

f In

form

atio

n T

heo

ryEle

men

ts o

f In

form

atio

n T

heo

ry

•The

cap

acity

of t

he c

hann

el•T

he c

ondi

tiona

l ent

ropy

func

tion

for e

ach

b k

•The

ave

rage

d va

lue

over

all

b k(e

quiv

ocat

ion)

•The

mut

ual i

nfor

mat

ion

of z

and

v

•The

cap

acity

∑=

−=

J jk

jk

jk

ba

Pb

aP

bH

1)

|(

log

)|

()

|(z

∑∑

∑=

==

−=

=J j

K kk

jk

jK k

kk

ba

Pb

aP

bP

bH

H1

11

)|

(lo

g)

,(

)(

)|

()

|(

zv

z

∑∑

∑∑

∑=

=

=

==

==

−=

J j

K kJ j

kjj

kjkj

jJ j

K kk

j

kj

kj

qa

Pqq

aP

bP

aP

ba

Pb

aP

HH

I1

11

11

)(

log

)(

)(

)(

),

(lo

g)

,(

)|

()

()

,(

vz

zv

z

[])

,(

max

vz

zI

C=

Page 15: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Fundam

enta

l Codin

g T

heo

rem

sFu

ndam

enta

l Codin

g T

heo

rem

s

•The

nois

eles

s cod

ing

theo

rem

: how

to c

ompa

ct in

form

atio

n as

muc

h as

pos

sibl

e•C

onsi

der a

sour

ce o

f inf

orm

atio

n w

ith fi

nite

ens

embl

e (A

,z) a

nd st

atis

tical

ly in

depe

nden

t so

urce

sym

bols

: so-

calle

d ze

ro-m

emor

y so

urce

•Non

-ext

ende

d so

urce

pro

duce

s one

sym

bol a

nd th

e n-

thex

tens

ion

prod

uces

blo

ck ra

ndom

va

riabl

e of

nsy

mbo

ls. I

ts e

ntro

py is

ntim

es th

e en

tropy

of t

he si

ngle

sym

bol s

ourc

e

•The

sour

ce o

utpu

t is a

n n-

tupl

eof

sym

bols

•The

fina

l res

ult i

s

w

here

L` a

vgis

the

aver

age

wor

d le

ngth

of t

he

code

cor

resp

ondi

ng to

the

nth

exte

nsio

n

•The

eff

icie

ncy

of a

ny e

ncod

ing

stra

tegy

is n

HnL

Hav

g1

)(

')

(+

<≤

zz

avg

LHn

')

(z=

η

Page 16: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Fundam

enta

l Codin

g T

heo

rem

sFu

ndam

enta

l Codin

g T

heo

rem

s

•The

nois

y co

ding

theo

rem

: how

to e

nsur

e a

relia

ble

com

mun

icat

ion

•The

sim

ples

t way

: by

a re

petit

ive

mes

sagi

ng

•Cod

e of

size

ϕan

d bl

ock

leng

th r

has r

ate

R=

log

(ϕ/r

)

•For

any

rate

R <

C, t

he c

apac

ity o

f the

cha

nnel

, the

re e

xist

an

inte

ger r

and

code

of b

lock

le

ngth

ran

d ra

te R

such

that

the

prob

abili

ty o

f blo

ck d

ecod

ing

erro

r is l

ess t

han

(any

) ε >

0.

•The

sour

ce c

odin

g th

eore

m: t

he c

hann

el is

err

or-f

ree

but t

he e

ncod

ing

proc

ess i

s los

sy,

henc

e a

rate

-dis

torti

on c

ompr

omis

e

Page 17: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Err

or-

free

Com

pre

ssio

nErr

or-

free

Com

pre

ssio

n

•Var

iabl

e-le

ngth

cod

ing

redu

ces o

nly

codi

ng re

dund

ancy

by

assi

gnin

g th

e sh

orte

st p

ossi

ble

code

wor

ds to

the

mos

t pro

babl

e (g

rey)

leve

ls.

•Huf

fman

cod

ing

(cod

es o

ne sy

mbo

l at a

tim

e)

•Ord

er th

e pr

obab

ilitie

s of t

he sy

mbo

ls

and

com

bine

the

low

est p

roba

bilit

y sy

mbo

ls in

to a

sing

le sy

mbo

l

•Cod

e ea

ch re

duce

d so

urce

wor

king

bac

k to

the

orig

inal

sour

ce.

0.97

3ef

f.

code

lbi

ts/s

ymbo

14.2

lbi

ts/s

ymbo

2.2

==

=

EL avg

Page 18: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Err

or-

free

Com

pre

ssio

nErr

or-

free

Com

pre

ssio

n

•For

a la

rge

num

ber o

f sym

bols

to b

e co

ded

Huf

fman

cod

ing

is c

ompu

tatio

nally

com

plex

. O

ther

cod

es p

rovi

de a

trad

e-of

f bet

wee

n co

ding

eff

icie

ncy

and

sim

plic

ity.

•Tru

ncat

ed H

uffm

an c

odin

g: c

odes

with

Huf

fman

cod

e on

ly th

e m

ost p

roba

ble

sym

bols

. All

othe

rs a

re c

oded

by

pref

ix a

nd fi

xed-

leng

th c

ode

•B-c

ode:

mad

e by

con

tinua

tion

bits

and

info

rmat

ion

bits

•Shi

ft co

des:

•Arr

angi

ng th

e sy

mbo

ls b

y th

eir (

mon

oliti

caly

dec

reas

ing)

pro

babi

litie

s

•div

idin

g th

e to

tal n

umbe

r of s

ymbo

ls in

to e

qual

ly si

zed

bloc

ks

•cod

ing

the

indi

vidu

al e

lem

ents

with

in a

ll bl

ocks

iden

tical

ly

•add

ing

shift

-up

and/

or sh

ift-d

own

sym

bols

to id

entif

y ea

ch b

lock

.

Page 19: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Err

or-

free

Com

pre

ssio

nErr

or-

free

Com

pre

ssio

n

Page 20: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Err

or-

free

Com

pre

ssio

nErr

or-

free

Com

pre

ssio

n

•Arit

hmet

ic c

odin

g

•An

entir

e se

quen

ce o

f sou

rce

sym

bols

(a m

essa

ge) i

s ass

igne

d a

sing

le a

rithm

etic

co

de w

ord

•The

cod

e w

ord

defin

es a

n in

terv

al o

f rea

l num

bers

bet

wee

n 0

and

1

•Eac

h sy

mbo

l red

uces

the

size

of t

he in

terv

al in

acc

orda

nce

with

its p

roba

bilit

y.

•Exa

mpl

e: fi

ve-s

ymbo

l mes

sage

from

a fo

ur-s

ymbo

l sou

rce

is b

eing

cod

ed

Page 21: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Err

or-

free

Com

pre

ssio

n:

Bit-p

lane

Codin

gErr

or-

free

Com

pre

ssio

n:

Bit-p

lane

Codin

g

•Bit-

plan

e de

com

posi

tion:

•Dire

ct se

para

tion

into

m1-

bit p

lane

s

•Gra

y co

de (s

ucce

ssiv

e co

de w

ords

diff

er in

onl

y on

e bi

t pos

ition

)

00

11

22

11

22

22

aa

aa

mm

mm

++⋅⋅⋅

++

−−

−−

11

12

0

−−

+

=−

≤≤

⊕=

mm

ii

i

ag

mi

aa

g

Exam

ple:

two

1024

×102

4 im

ages

: one

mon

ochr

ome

a on

e bi

nary

Page 22: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Err

or-

free

Com

pre

ssio

n:

Bit-p

lane

Codin

gErr

or-

free

Com

pre

ssio

n:

Bit-p

lane

Codin

g

Page 23: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Err

or-

free

Com

pre

ssio

n:

Bit-p

lane

Codin

gErr

or-

free

Com

pre

ssio

n:

Bit-p

lane

Codin

g

Page 24: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Err

or-

free

Com

pre

ssio

n:

Bit-p

lane

Codin

gErr

or-

free

Com

pre

ssio

n:

Bit-p

lane

Codin

g

•Con

stan

t Are

a C

odin

g (C

AS)

•The

imag

e is

div

ided

into

blo

cks o

f siz

e p×

q, c

lass

ified

as w

hite

, bla

ck o

r mix

ed; t

he m

ost

prob

able

is a

ssig

ned

0, th

e ot

her t

wo

cate

gorie

s are

ass

igne

d 10

and

11.

•Whi

te b

lock

skip

ping

(WB

S): w

hite

blo

cks a

re c

oded

by

0 an

d al

loth

ers b

y a

1 fo

llow

ed b

y th

e bi

t pat

tern

of t

he b

lock

•O

ne-d

imen

sion

al ru

n-le

ngth

cod

ing

•Rep

rese

nt (r

ow-w

ise)

sucs

essi

veru

ns o

f bla

ck o

r whi

te p

ixel

s•T

he ru

n-le

ngth

s can

be

code

d ad

ditio

nally

by

a va

riabl

e-le

ngth

cod

e•T

wo-

dim

ensi

onal

run-

leng

th c

odin

g

Page 25: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Loss

less

Pre

dic

tive

Codin

gLo

ssle

ss P

redic

tive

Codin

g

Page 26: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Loss

less

Pre

dic

tive

Codin

gLo

ssle

ss P

redic

tive

Codin

g

Page 27: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Loss

yCom

pre

ssio

nLo

ssy

Com

pre

ssio

n

•Los

syen

codi

ng is

bas

ed o

n th

e co

ncep

t of c

ompr

omis

ing

the

accu

racy

of t

he re

cons

truct

ed

imag

e in

exc

hang

e fo

r inc

reas

ed c

ompr

essi

on

•Qua

ntiz

atio

n bl

ock

is a

n es

sent

ial p

art o

f the

enc

oder

•Los

sypr

edic

tive

codi

ng

•The

qua

ntiz

erm

aps t

he p

redi

ctio

n er

ror i

nto

a lim

ited

rang

e of

out

puts

•The

pre

dict

or is

pla

ced

in th

e fe

edba

ck lo

op, t

hus a

chie

ving

the

sam

e pr

edic

tion

bloc

k in

the

deco

der

Page 28: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Loss

yCom

pre

ssio

nLo

ssy

Com

pre

ssio

n

•Del

ta m

odul

atio

n: a

n on

e-bi

t per

pix

el re

pres

enta

tion

•Exa

mpl

e fo

r α =

1,

ζ =

6.5

•Eff

ects

such

as g

ranu

lar n

oise

(αto

o hi

gh)

or sl

ope

over

load

(αto

o sm

all)

othe

rwis

e0

for

ˆ

1

>

−+=

=−

nn

nn

ee

ff

ζζα

&&

Page 29: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Loss

yCom

pre

ssio

nLo

ssy

Com

pre

ssio

n

•Diff

eren

tial p

ulse

cod

e m

odul

atio

n:

an o

ptim

al in

mea

n-sq

uare

sens

e pr

edic

tor.

•Con

stra

ints

:

•qua

ntiz

atio

n er

ror a

ssum

ed to

be

negl

igib

le

•the

pre

dict

ion

is a

line

ar c

ombi

natio

n of

m p

revi

ous p

ixel

s.

•Com

puta

tion

of th

e au

toco

rrel

atio

n m

atrix

is d

iffic

ult i

n pr

actic

e. In

stea

d, a

m

odel

(e.g

. 2-D

Mar

kov

sour

ce) i

s us

ed

•The

sum

of t

he p

redi

ctio

n co

effic

ient

s sh

ould

be

smal

ler t

han

1 to

ens

ure

the

sam

e ou

tput

gre

y le

vels

{}

[]

{}

[]

[]

{}

{}

{}

{}

{}

{}

{}

{}

{}

{}

=

=

⋅⋅

⋅⋅

⋅=

=

=−

=

=

=+

≈+

=

−−−

−−

−−

−−

−−

−−

−−

−−

×

=−

=−

∑∑

mm

nn

nn

nn

mn

mn

nm

nn

mn

nn

mn

nn

nn

n

mm

m ii

ni

nn

nn

m ii

ni

n

nn

nn

nn

ff

E

ff

Ef

fE

ff

Ef

fE

ff

E

ff

Ef

fE

ff

Ef

fE

ff

Ef

fE

eE

ff

ff

ef

ef

ααα

α

α

α

α

:

;

:

.........

w

here

ˆ

ˆ

ˆˆ

221

211

2

12

11

1

1

2

1

22

1

1

r

RrR&

&

Page 30: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Loss

yCom

pre

ssio

nLo

ssy

Com

pre

ssio

n

•Opt

imal

qua

ntiz

atio

n: g

iven

a

stai

rcas

e fu

nctio

n t =

q(s

), th

e de

sign

pr

oble

m is

to se

lect

the

best

s iand

t ifor

an

inpu

t den

sity

func

tion

p(s)

•The

reco

nstru

ctio

n le

vels

t iare

th

e ce

ntro

idso

f the

are

as u

nder

p(

s)

•The

dec

isio

n le

vels

s iare

ha

lfway

bet

wee

n th

e re

cons

truct

ion

leve

ls

•The

obt

aine

d qu

antiz

eris

cal

led

(L-le

vel)

Lloy

d-M

ax q

uant

izer

2/1

2/,..

.2,1

0

20

2/,..

.,2,1

0)

()

(

1

1

Li

Li

it

ts

Li

sp

ts

ii

is si

i i

=−

==

∞+=

==

+

−∫

Page 31: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g

•Gen

eral

sche

me

•The

tran

sfor

m h

as to

dec

orre

late

the

pixe

ls o

r to

com

pact

as m

uch

info

rmat

ion

as

poss

ible

into

the

smal

lest

num

ber o

f tra

nsfo

rm c

oeff

icie

nts

•The

qua

ntiz

atio

n se

lect

ivel

y el

imin

ates

or m

ore

coar

sely

qua

ntiz

es th

e le

ss

info

rmat

ive

coef

ficie

nts

•Var

iabl

e-le

ngth

cod

ing

elim

inat

es th

e re

mai

ned

codi

ng re

dund

ancy

Page 32: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g

•Tra

nsfo

rm se

lect

ion

•Def

ined

by

forw

ard

and

inve

rse

trans

form

atio

n ke

rnel

s (ba

sis

func

tions

, bas

is im

ages

)

•Sep

arab

le k

erne

ls

•Sym

met

ric k

erne

ls)

,(

),

()

,,

,(

),

()

,(

),

,,

(

1,..

.2,1,0

,fo

r x

),

,,

()

,(

),

(

1,..

.2,1,0

,fo

r )

,,

,(

),

()

,(

11

211 0

1 0

1 0

1 0

vy

gu

xg

vu

yx

gv

yg

ux

gv

uy

xg

Ny

vu

yx

hv

uT

yx

f

Nv

uv

uy

xg

yx

fv

uT

N u

N v

N x

N y

==

−=

=

−=

=

∑∑

∑∑

− =

− =

− =

− =

•Fou

rier t

rans

form

•Wal

sh-H

adam

ard

trans

form

Nyv

uxj

Nyv

uxj

ev

uy

xh

eN

vu

yx

g

/)(

2

/)(

22

),

,,

(

1)

,,

,(

+

+−

==

π

π

)(

)(

)(

:)

()

()

()

()

()

()

()

(

2

whe

re)1

(1

),

,,

()

,,

,(

01

1

32

2

21

1

10

)(

)(

)(

)(

0

ub

ub

up

ub

ub

up

ub

ub

up

ub

up

NN

vu

yx

hv

uy

xg m

mm

mmm

mv

py

bu

px

bm i

ii

ii

+=

+=

+==

=∑

−=

=

−−

−−−

+=

Page 33: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g

•Wal

sh-H

adam

ard

trans

form

ker

nel c

onsi

sts o

f alte

rnat

ing

plus

and

min

us 1

’s a

rran

ged

in

a ch

ecke

rboa

rd p

atte

rn

Page 34: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g 1,..

.2,1

for

0fo

r

21

)(

2)1

2(co

s2

)12(

cos

)(

)(

),

,,

()

,,

,(

−==

=

+

+

==

Nuu

NNu

Nv

yN

ux

vu

vu

yx

hv

uy

xg α

ππ

αα

•Dis

cret

e C

osin

e Tr

ansf

orm

Page 35: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g

50%

of t

rans

form

coe

ffic

ient

s dis

rega

rded

Res

idua

l im

ages

are

not

zer

o bu

t som

e m

ean-

squa

re e

rror

rms Fo

urie

r=1.

28

rms W

alsh

=0.8

6

rms D

CT=

0.68

Page 36: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g

[]

∑∑

∑∑

∑∑∑

− =

− =

− =

− =

− =

− =− =

− =

−=

−==

=

−−

−−

⋅⋅

⋅⋅

⋅−

=

=

=

1 0

1 02

),

(

2

1 0

1 0

1 0

1 01 0

1 0

),

(1

ˆ

),

()

,(

ˆt

sign

ifica

n is )

,(

if

tsi

gnifi

can

-no

n is )

,(

if 10

),

(

),

,1,1

(...

),

,1,1(

),

,0,1(

......)

,,0,1(

),

,1,0(

...)

,,1,0(

),

,0,0(

),

(

),

,,

()

,(

),

(

n u

n vv

uT

ms

n u

n vuv

uv

n u

n vuv

n u

n v

vu

Ee

vu

Tv

u

vu

Tv

uT

vu

vu

nn

hv

un

hv

un

h

vu

hv

un

hv

uh

vu

h

vu

T

vu

yx

hv

uT

yx

f

γσ

γ

γ

FF

HFH

HF

•Bas

is im

ages

and

ene

rgy

com

pact

ion

•The

mat

rix F

is d

efin

ed a

s a li

near

co

mbi

natio

n of

n2

mat

rices

of s

ize

n×n

(bas

is m

atric

es)

•mas

king

func

tion

to g

et tr

unca

ted

expa

nsio

n

•The

mea

n-sq

uare

app

roxi

mat

ion

erro

r is t

he su

m o

f the

var

ianc

es o

f th

e di

scar

ded

coef

ficie

nts

•The

Opt

imal

Kar

hune

n-Lo

eve

Tran

sfor

m a

nd it

s app

roxi

mat

ions

Page 37: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g

•Bou

ndar

y (G

ibbs

-like

) eff

ects

: mor

e vi

sibl

e fo

r the

Fou

rier t

rans

form

and

less

vis

ible

for

the

DC

T

Page 38: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g

•Sub

imag

esi

ze se

lect

ion:

com

puta

tiona

l com

plex

ity v

ersu

s cod

ing

effic

ienc

y

Page 39: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

codin

gTra

nsf

orm

codin

g

Page 40: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g

•Bit

allo

catio

n: z

onal

cod

ing

(bas

ed o

n m

axim

um v

aria

nce)

or

thre

shol

d co

ding

(bas

ed o

n m

axim

um a

mpl

itude

)

•Zon

al c

odin

gim

plem

enta

tion:

info

rmat

ion

is v

iew

ed a

s un

certa

inty

. Tra

nsfo

rm c

oeff

icie

nts w

ith m

axim

um v

aria

nce

carr

y te

mos

t of i

mag

e in

form

atio

n an

d ha

ve to

be

reta

ined

.

Page 41: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g

•Zon

al c

odin

gim

plem

enta

tion:

info

rmat

ion

is v

iew

ed a

s unc

erta

inty

. Tra

nsfo

rmco

effic

ient

s with

max

imum

var

ianc

e ca

rry

tem

ost o

f im

age

info

rmat

ion

and

have

to b

e re

tain

ed.

•Zon

al m

asks

are

show

ing

the

num

ber o

f bits

use

d to

cod

e ea

ch c

oeff

icie

nt

•Fix

ed n

umbe

r of b

its fo

r eac

h co

eff.

•Bit

budg

et d

istri

bute

d un

equa

lly

•Llo

yd-M

ax q

uant

izer

•Ray

leig

hpd

fmod

el fo

r the

dc

•Lap

laci

anfo

r the

rem

aini

ng c

oeff

.

Page 42: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g

•Thr

esho

ld c

odin

gim

plem

enta

tion:

Tra

nsfo

rm c

oeff

icie

nts o

f lar

gest

mag

nitu

de m

ake

to

mos

t sig

nific

ant c

ontri

butio

n to

reco

nstru

cted

subi

mag

equ

ality

. •T

he re

tain

ed c

oeff

icie

nts a

re re

orde

red

in 1

-D m

anne

r (zi

g-za

gsc

anni

ng) a

nd th

en ru

n-le

ngth

cod

ed

•Thr

esho

ldin

g •S

ingl

e gl

obal

thre

shol

d to

all

coef

f.

•lev

el o

f com

pres

sion

var

ies

•Diff

eren

t for

diff

eren

t sub

imag

es

•con

stan

t cod

ing

rate

•Var

ying

(with

the

loca

tion)

thre

shol

d

•com

bine

s thr

esho

ldin

g w

ith

qu

antiz

atio

n

Page 43: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g

),

()

,(ˆ

),

(

),

()

,(

),

)1,1

(...

)1,1(

)0,1(

......)0,1(

)1,0(

...)1,0(

)0,0(

vu

Zv

uT

vu

T

vu

Zv

uT

roun

dv

uT

nn

Zn

Zn

Z

Zn

ZZ

Z =

=

−−

−−

⋅⋅

⋅⋅

⋅−

=

&ZZ

is a

tran

sfor

m n

orm

aliz

atio

n ar

ray

Bef

ore

rest

orat

ion,

the

norm

aliz

ed (t

hres

hold

ed a

nd

quan

tized

) im

age

mus

t be

deno

rmal

ized

.

Page 44: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Tra

nsf

orm

Codin

gTra

nsf

orm

Codin

g Cr1

=34

Cr2

=67

Page 45: No Slide Titleagotchev/DIPII/presentation3.pdf · Fundamentals: Interpixel Fundamentals: Redundancy Interpixel Redundancy. Example of a non-visual mapping: •1024x343 grey-scale

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

Dig

ital I

mag

e Pro

cess

ing,

2nd

ed.

www.imageprocessingbook.com

© 2

002

R. C

. Gon

zale

z &

R. E

. Woo

ds

Imag

e Com

pre

ssio

nIm

age

Com

pre

ssio

n

End o

f Par

t 1

End o

f Par

t 1