Scribd
Upload
261 views

Laplacian in Image Processing

1. The document discusses techniques for image enhancement in the spatial domain, including histogram equalization, averaging noisy images, calculating the discrete Laplace operator, and relating the subtraction of the Laplacian to unsharp masking. 2. It provides an example of equalizing the histogram of an 8x8 image, and averages multiple noisy versions of an image to reduce noise by taking the mean. 3. It derives the discrete Laplace operator, compares it to the continuous case, and shows that subtracting the Laplacian is proportional to unsharp masking.

Uploaded by

Phan Thế Duy
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
261 views

Laplacian in Image Processing

1. The document discusses techniques for image enhancement in the spatial domain, including histogram equalization, averaging noisy images, calculating the discrete Laplace operator, and relating the subtraction of the Laplacian to unsharp masking. 2. It provides an example of equalizing the histogram of an 8x8 image, and averages multiple noisy versions of an image to reduce noise by taking the mean. 3. It derives the discrete Laplace operator, compares it to the continuous case, and shows that subtracting the Laplacian is proportional to unsharp masking.

Uploaded by

Phan Thế Duy
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Exercise 2/06

51

T-61.5100 Digital image processing, Exercise 2/06


Image enhancement in the spatial domain
1. Equalize the histogram of the 8 8 image below. The image has grey levels 0, 1, . . . , 7.

4
4
4
4
4
4
4
4

4
5
5
5
5
5
4
4

4
5
6
6
6
5
4
4

4
5
6
7
6
5
4
4

4
5
6
6
6
5
4
4

4
5
5
5
5
5
4
4

4
4
4
4
4
4
4
4

0
0
0
0
0
0
0
0

2. Assume that we have many noisy versions gi (x, y) of the same image f (x, y), i.e.
gi (x, y) = f (x, y) + i (x, y)
where the noise i is zero-mean and all point-pairs (x, y) are uncorrelated. Then we can
reduce noise by taking the mean of all the noisy images
M
1 X
gi (x, y).
g(x, y) =
M
i=1

Prove that
E{
g (x, y)} = f (x, y)
and
g2(x,y) =

1 2

M (x,y)

2
where (x,y)
is the variance of and g2(x,y) the variance of g(x, y).

3. A two-variable continuous functions Laplace-operator is


52 f (x, y) =

2f
2f
+
.
x2
y 2

From this, it follows that


F{52 f (x, y)} = (2)2 (u2 + v 2 )F (u, v),
where F (u, v) = F{f (x, y)}. Determine the corresponding operator and the Fourier transform in the discrete case. Compare the result obtained to the continuous case.
4. Show that subtracting the Laplacian from an image is proportional to unsharp masking.
Use the definition for the Laplacian in the discrete case.

52

Exercise 2/06

T-61.5100 Digital image processing, Exercise 2/06


1.
Histogram equalization enhances contrast in an image (more gray levels are used). It is usually
used to improve the visibility of details in images. The discrete equalization is given by
sk = T (rk ) =

k
X
nj
j=0

where rk is the normalized gray level, nk is the number of pixels having gray level k, and n is
the total number of pixels.
First we calculate the gray level histogram:
k
rk
nk
pr (rk ) = nk /n

0
0
8
0.125

1
0.143
0
0

2
0.286
0
0

3
0.429
0
0

4
0.571
31
0.484

5
0.714
16
0.250

6
0.857
8
0.125

7
1.0
1
0.016

Then we form the cumulative distribution function:


sk

0.125

0.125

0.125

0.125

0.609

0.859

0.984

1.000

Finally we round the obtained values for the actual gray levels rk :
s0k

0.143

0.143

0.143

0.143

0.571

0.857

1.000

1.000

So, the histogram equalization is done by mapping the gray levels k into the new gray levels k 0 :
k
k0

0
1

1
1

2
1

3
1

4
4

5
6

6
7

7
7

Resulting image:

4
4
4
4
4
4
4
4

4
6
6
6
6
6
4
4

4
6
7
7
7
6
4
4

4
6
7
7
7
6
4
4

4
6
7
7
7
6
4
4

4
6
6
6
6
6
4
4

4
4
4
4
4
4
4
4

2.
The noisy image is now given as
gi (x, y) =
f (x, y)
+ i (x, y),
| {z }
| {z }
| {z }
noisy image original image
noise
where noise has zero mean and it is uncorrelated.

1
1
1
1
1
1
1
1

Exercise 2/06

53

When we average the noisy images


g(x, y) =

M
1 X
gi (x, y)
M
i=1

we get for the expectation


(
)
(
)
M
M
M
1 X
1 X
1 X
E {
g (x, y)} = E
gi (x, y) = E
f (x, y) +
i (x, y) =
M
M
M
i=1

i=1

i=1

M
M
1 X
1 X
E {f (x, y)} +
M
M
i=1

i=1

E {i (x, y)} = f (x, y)


|
{z
}
=0 (zero mean)

and for the variance

!2
!2
M
M

1 X

1 X
n
o
=E
=
g2(x,y) = E (
g E {
g })2 = E
(f + i ) f
i

M
i=1
i=1

!2
M
M
M
M
M

X
X
X
X
X
1
1

2
i2 +
= 1
E{
}
+
E{i j }
E

i
i
j
i
| {z }
M2
M2
M2
| {z }
i=1

i=1

i=1

j=1, j6=i

=2

j=1, j6=i

=0

1 2
,
M

since the noise had zero mean and it was uncorrelated.

3.
We approximate the derivative in discrete case:
f (x, y)
' f (x, y) f (x 1, y)
x
By derivating again, we have
2 f (x, y)
' (f (x, y)f (x1, y))(f (x1, y)f (x2, y)) = f (x, y)2f (x1, y)+f (x2, y).
x2
This is equivalent to filtering with a mask [1, 2, 1]. The mask is most practically used symmetrically so that 2 is in the middle, rather than as in the formula above.
Thus
2 f (x, y) 2 f (x, y)
+
' f (x 1, y) + f (x + 1, y) + f (x, y 1) + f (x, y + 1) 4f (x, y),
x2
y 2
which corresponds to mask h(x, y):
1

1
4
1

Next we calculate the Fourier transform of the mask:


H(u, v) =

1
1 X X
h(x, y)ej2(ux+vy)/N = (ej2u/N + ej2v/N + ej2u/N + ej2v/N 4)
N x= y=
N

1
[2 cos(2u/N ) + 2 cos(2v/N ) 4]
N

54

Exercise 2/06

And finally
1
(2 cos(2u/N ) + 2 cos(2v/N ) 4) F (u, v),
N

F{52 f [x, y]} =

10
9
8
7
6
5
4
3
2
1
0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

In the figure there is |H(u, 0)| in discrete and continuous case. Solid line: discrete case, i.e.
|H(u, 0)| = 2 2 cos(2u). Dotted line: continuous case, i.e. |H(u, 0)| = (2)2 u2 .

4.
Consider the following equation:
f (x, y) 52 f (x, y) = f (x, y) [f (x + 1, y) + f (x 1, y) + f (x, y + 1) + f (x, y 1) 4f (x, y)]
= 6f (x, y) [f (x + 1, y) + f (x 1, y) + f (x, y + 1) + f (x, y 1) + f (x, y)]
1
= 5{1.2f (x, y) [f (x + 1, y) + f (x 1, y) + f (x, y + 1) + f (x, y 1) + f (x, y)]}
5

= 5[1.2f (x, y) f (x, y)]


where f(x, y) denotes the average of f (x, y) in a predefined neighborhood that is centered at
(x, y) and includes the center pixel and its four immediate neighbors. Treating the constants in
the last line of the above equation as proportionality factors, we may write
f (x, y) 52 f (x, y) f (x, y) f(x, y).
The right side of this equation is recognized as the definition of unsharp masking given in
Eq. (3.7-7). Thus, it has been demonstrated that subtracting the Laplacian from an image is
proportional to unsharp masking.

You might also like