SCALAR QUANTISATION Let

represent a continuous scalar quantity which

could be one of the following: pixel intensity transform coefficient image model parameter other levels are used to represent .

Suppose that only

This process is called amplitude quantisation. Scalar quantisation Each scalar is quantised independently. Vector quantisation Two or more scalars are quantised jointly, i.e., the vector formed by two or more scalars is quantised. Let
f

denote an

that has been quantised.

11

 = Q ( f ) = ri f

, d i1 < f d i

: quantisation operation : the :

L +1 L

r i di

reconstruction levels, 1 i L

decision boundaries, 0 i L

can be expressed as
= Q ( f ) = f + eQ f

, where e

Q(

f f)= f

eQ

: quantisation noise

f ) (f
f ) = f f d( f

d( f

1/ p

The reconstruction and decision levels are often determined by minimizing some error criterion based on
d ( f , ) f

12

Example:

Minimise the average distortion

] = D = E [d ( f , f f ) p f ( f 0 )df 0 d ( f0 , f

D,

defined as follows:
0=

Uniform quantisation In uniform quantisation the reconstruction and decision levels are uniformly spaced.
di d i 1 = , 1 i L
d i + d i 1 , 1 i L 2

and

ri =

f
7 ( r4 ) 8
5 ( r3 ) 8

3 ( r2 ) 8
1 ( r1 ) 8

0( d 0 ) 1 ( d 1 )
4

1 (d 2 ) 2

3 ( d 3 ) 1( d 4 ) 4

13

Figure 1.1: Example of uniform quantisation. The number of reconstruction levels is 4, assumed to be between 0 and 1, and result of quantising respectively.
eQ = f f
1/8 1/4 1/8 1/2 3/4 1
f

is

is the and
di

. The reconstruction levels

ri

and decision boundaries are denoted by

14

Figure 1.2: Illustration of

signal

dependence

of

quantisation noise in uniform quantisation Uniform quantisation may not be optimal! Suppose
f

is much more likely to be in one particular

region that in others. It is reasonable to assign more reconstruction levels to that region! Quantisation in which reconstruction and decision levels do not have even spacing is called non-uniform quantisation. Optimum determination of error criterion used. Quantisation using the MMSE criterion Suppose
f

r i

and

di

depends on the

is a random variable with a pdf

p f ( f0 )

15

Using the minimum mean squared error (MMSE) criterion, we determine average distortion
= f

rk

and

dk

by minimising the

given by
f 0 ) 2 df 0 p f ( f 0 )( f

2 )] = E [eQ f )2 ] D = E [d ( f , f ] = E [( f
0 =

Noting that write

is one of the
L

reconstruction levels we

D = f i =d
i =1
0

i 1

p f ( f 0 )( ri f 0 ) 2 df 0

To minimize

D =0 rk

, ,

1 k L

D =0 d k

1 k L 1

d 0 =

dL =

It is proven that from the above we get

rk =

f0 =d k 1 f 0 p f ( f 0 )df 0 f0 =d k 1 p f ( f 0 )df 0
rk + rk + 1 2
dk

dk

1 k L

dk =

1 k L 1

d 0 =

dL =

Note that:
16

 The reconstruction level over the interval

d k 1 f 0 d k dk

rk

is the centroid of and

p f ( f0 )

.
d0

The decision level

except

dL

is the middle
rk

point between two reconstruction levels equations for the optimal solution.

and

rk +1

The above set of equations is a necessary set of For a certain class of pdf's including uniform, Gaussian, Laplacian, is also sufficient. A quantiser based on the MMSE criterion is often referred to as Lloyd-Max quantiser.
f

1.5104

0.4528 -0.9816 -0.4528 0.9816

-1.5104

17

Figure 1.3: Example of a Lloyd-Max quantiser. The number of reconstruction levels is 4, and the probability for variance 1.
f

is Gaussian with mean 0 and

2 VECTOR QUANTISATION Let

fi
f = [ f1 , f 2 , , f k ]T

denote an

k -dimensional

vector that

consists of .

real-valued, continuous-amplitude scalars

k -dimensional

is mapped to another is chosen from

N

vector

y = [ y1 , y 2 , , y k ]T

possible reconstruction or

quantisation levels
= VQ( f ) = y , f C f i i

VQ is the vector quantisation operation

Ci

is called the ith cell.

18

distortion measure: quantisation noise:

, f ) = eT e d (f Q Q
f = VQ( f ) f eQ = f

PROBLEM:determine SOLUTION: by
)] D =E [d ( f , f

yi

and boundaries of cells Ci

D

minimise some error criterion such given

as the average distortion measure

T -f ) T ( f -f ) p ( f ) df D = E [e T ( f Q e Q ] = E [( f-f ) ( f-f )] = 0 0 f 0 0 -

= (ri -f0 ) T (ri -f0 ) p f ( f0 ) df0

i= 1 f0Ci

MAJOR source. That means

ADVANTAGE

OF

VQ:

performance

improvement over scalar quantisation of a vector

VQ can lower the average distortion

with the

number of reconstruction values held constant. VQ can reduce the required number of reconstruction values when
D

is held constant.

19

f2

f1

Figure 2.1: Example of vector quantisation. The number of scalars in the vector is 2, and the number of reconstruction levels is 9.

20