18CSE357T – BIOMETRICS

Unit –2 : Session –2 : SLO -1

SRM Institute of Science and Technology 1

• Objectives of image pre-processing:
(a) Enhancing information that is useful for later
analysis

(b) Suppress image information that is not relevant

to later work

5-2
• Classes of Image Pre-processing Methods
(a) Brightness Transformations
(b) Geometric transformations

5.1 Brightness Transformations

Categorization:
(1) Point processing, Neighborhood processing
(2) Position invariant, Position variant
(3) Image Enhancement, Image Restoration

5-3
○ Point Processing
• Histogram Equalization (HE)

5-4
Transform function
x−a 
y=( ) ( d − c ) + c,
b−a
a xb
14 − 2
e.g., y = ( x − 5) + 2
9−5
5 x9
5-5
  1
 =1  =2  = −2

5-6
Theorem: Let T be a differentiable strictly increasing
or strictly decreasing function.
Let r be a random variable having density pr
Let s = T ( r ) having density ps
dr
Then, ps ( s ) = pr ( r ) or ps ( s ) ds = pr (r ) dr
ds

5-7
Proof: Let Pr ,Ps : the distribution functions of r and s
(a) T strictly increasing
Ps (s) = P( s  s) = P(T (r )  s) = P(r  T −1 (s)) = Pr (T −1 (s))
−1 −1 −1
 dP (T ( s ))  dT ( s ) dT (s)
Ps ( s ) = r
= Pr (T ( s ))
−1
= pr (T ( s ))
−1

ds ds ds
−1
dT −1 ( s ) dT −1 ( s ) −1 dT ( s)
= , ps ( s) = pr (T ( s))
ds ds ds
(b) T strictly decreasing
Ps (s) = P( s  s) = P(T (r )  s) = P(r  T −1 (s)) = 1 − Pr (T −1 (s))
−1 −1 −1
 dP (T ( s ))  dT ( s ) dT (s)
Ps ( s ) = r
= − Pr (T ( s ))
−1
= pr (T ( s ))(−
−1
)
ds ds ds
−1
dT −1 ( s ) dT −1 ( s ) −1 dT ( s)
− = , ps ( s) = pr (T ( s))
ds ds ds 5-8
Let transform function T (r ) be s = T (r ) = r p (w)dw
r
0

Then ds = pr (r )
dr
dr 1
ps (s ) = pr (r )| |,  ps (s ) = pr (r )| | =1
ds pr (r )

Called equalization or linearization.

5-9
○ Example

−r + 1 0  r  1
Let p r (r ) = 
 0 elsewhere

r 1 2
Since s = T (r ) =  ( − w + 1)dw = − r + r
0 2
the transform function T is

5-10
 1 2
s = − r + r r = T (s )=1  1 − 2 s
−1

2 dr 1
Since 0  r  1, r = 1 − 1 − 2 s , =
ds 1 − 2s
From dr 1
ps (s ) = pr (r )| | = ( − r + 1)
ds 1 − 2s
1
= ( − 1+ 1 − 2 s +1)
1 − 2s
1
= 1 − 2s =1
i.e., 1 − 2s
is a uniform distribution
ps (s)

5-11
Discrete case:

n, k
Let p(rk ) = 0  rk  1 k = 1,2,,L − 1
,
n
Transform: knj
sk = T (rk ) = 
j =0 n
k
nj
Scale sk : ( L − 1) 
j =0 n

5-12
5-13
○ Examples:

5-14
Specified Histogram Equalization (SHE)
-- Specify the shape of the histogram that we wish
the processed image to have.

Input image HE SHE

5-15
5-16
Procedure:
Given: input image (I) and specification ( pz )
1. Compute the probability density pr of gray
levels r of the input image I
r
2. Compute T (r ) =  pr (w)dw from pr
0

z
3. Compute G (z ) =  pz (t )dt from pz
0

4. Compute z = G −1 (T (r ))
5. Transform I into O by z = G (T (r ))
−1

z : gray levels of the output image O

5-17
Discrete case:
k k nj
sk = T (rk ) =  pr (rj ) =  , k = 1,2,,L − 1
j =0 j =0 n

k
G (zk ) =  pz (zi ) = sk , k = 1,2,,L − 1
i =0

zk = G (sk ) = G (T (rk ))
−1 −1

5-18
Example: Given image I of size 64 by 64 with 8
gray levels r0 , r1 , , r7
Histogram of input image I:

k k nj
Transformation function: sk = T (rk ) = 
j =0
pr (rj ) =  ,
j =0 n

5-19
Specified histogram:

k
Transformation function: G (zk ) =  pz (zi ) = sk ,
i =0

5-20
Inverse transformation function: zk = G −1 (T (rk )) = G −1 (sk )

Output image O: zk = G −1 (T (rk )) = G −1 (sk )

5-21
Histogram of output image O:

5-22
Input histogram Equalized histogram

Output histogram Specified histogram

5-23