Digital Image Processing

Image Enhancement
Differences
Introduction:
NEED FOR ENHANCEMENT
Differences
Image Quality Factors
Contrast

Contrast Refers to Finer details of Image

Contrast

Methods for Computing contrast of an Image

Contrast Manipulation
Brightness
Image Quality Assessment tool
Histograms as assessment tool
Advantages of Histogram
Dynamic Range
Histogram assessment
Image Quality Metrics
Image Quality Metrics
Example
 Intensity Transformation Function

Log Transformations
➢ The Log function has the important characteristic that it compresses the
dynamic range of images with large variation in the pixel value.
Classical example is displaying Fourier spectrum.
▪ Fourier spectrum has the values in the range 0 to 1.5x106. These values are
scaled linearly for the display in 8 bit system.

Fourier Log transformation

spectrum with c=1
 Intensity Transformation Function

Power-law (Gamma ) Transformations

This has the basic form s=c rγ ,where c and γ are positive constants

Fractional values of γ maps a narrow

range of dark input values into a wider
range of output values. Opposite of this
also true for higher values of input
levels.

These are also called as gamma

correction due to the exponent in the
power law equation.

Plot for c=1

 Intensity Transformation Function

Power-law (Gamma ) Transformations

CRT device have an intensity to

voltage response that is a power
function with exponent varying from
approximately 1.8 to 2.5. Such
display system would produce
images that are darker than intended.

Pre-processed image with

gamma correction
s=r =r0.4. before input to
1/2.5

the display device

 Intensity Transformation Function

➢Gamma correction is very important

when to reproduce an image exactly
on a display system.
➢ Power-law transformations are also
used in general purpose contrast
manipulation.
close all
clear all;
clc;
[filename, pathname] = uigetfile('*.tif');
im = imread([pathname filename]);
imshow(im);
im1=double(im).^0.3;
im1=mat2gray(im1);
figure,imshow(im1);
 Intensity Transformation Function

MRI of Result of a Result of a Result of a

fractured transforma transforma transforma
human tion for tion for tion for
spine γ=0.6 γ=0.4 γ=0.3
Intensity Transformation Function

Arial
Result of a
image
transformation for c=1
and γ=3
Intensity Transformation Function

Result of a Result of a
transformation transformation for
for c=1 and c=1 and γ=5
γ=4
 Piecewise Linear transformation functions

Contrast stretching
➢ Low contrast images result from the following
▪ Poor illumination
▪ lack of dynamic range in the imaging sensor
▪ Wrong settings of the lens aperture during acquisition

➢It is a process that expands the range of intensity levels in an image so that it
spans full intensity range of the recording medium or display device
 Piecewise Linear transformation functions

Contrast stretching
L-1
Controls the shape of the
(r2,s2) transformation function
output intensity level, s

3L/4

L/2 T(r)

L/4
(r1,s1)
0
0 L/4 L/2 3L/4 L-1
Input intensity level, r
 Piecewise Linear transformation functions

Contrast stretching
Suppose r1=s1 and r2=s2

L-1 L-1

(r2,s2)
output intensity level, s

output intensity level, s

3L/4 3L/4 (r2,= s2)

L/2 T(r) L/2 T(r)

L/4 L/4
(r1,=s1)
(r1,s1)
0 0
0 L/4 L/2 3L/4 L-1 0 L/4 L/2 3L/4 L-1
Input intensity level, r Input intensity level, r
 Piecewise Linear transformation functions

Contrast stretching
Suppose r1=r2 and s1=0 and s2=L-1

L-1 (r2)
L-1

(r2,s2)
output intensity level, s

output intensity level, s

3L/4 3L/4

L/2 T(r) L/2 T(r)

L/4 L/4
(r1,s1)
0 0 (r1)
0 L/4 L/2 3L/4 L-1 0 L/4 L/2 3L/4 L-1
Input intensity level, r Input intensity level, r
 Piecewise Linear transformation functions

Contrast stretching
Intermediate values of (r1,s1) and (r2,s2) produces various degree of spread in
the intensity
L-1

(r2,s2)
output intensity level, s

3L/4

T(r)
L/2

L/4
(r1,s1)
0
0 L/4 L/2 3L/4 L-1
Input intensity level, r
Piecewise Linear transformation functions

Contrast stretching (Example)

(r1,s1)=(rmin, 0) and (r2,s2)=(rmax,L-1)

output intensity level, s L-1

3L/4 rmax,L-1

L/2 T(r)

L/4 rmin, 0

0
0 L/4 L/2 3L/4 L-1
Input intensity level, r
Piecewise Linear transformation functions

Contrast stretching (Example)

POOR CONTRAST IMAGE CONTRAST STRETCHED IMAGE
Piecewise Linear
transformation functions
Contrast stretching (Example)
CONTRAST STRETCHED IMAGE
 Piecewise Linear transformation functions

Contrast stretching (Example MATLAB PROGRAM)

close all
clear all;
clc;
[filename, pathname] = uigetfile('*.tif');
im = imread([pathname filename]);
imshow(im);
title('POOR CONTRAST IMAGE');
J = imadjust(im,[0.2 0.5],[0 1]);
figure,imshow(J);
title('CONTRAST STRETCHED IMAGE');
Piecewise Linear transformation functions

Contrast stretching (Example)

close all
clear all;
clc;
[filename, pathname] = uigetfile('*.tif');
im = imread([pathname filename]);
imshow(im);
title('POOR CONTRAST IMAGE');

K=im2bw(im,0.42);
figure,imshow(K)
Piecewise Linear
transformation functions
Contrast stretching (Example)
POOR CONTRAST IMAGE
 Piecewise Linear transformation functions

Intensity Level slicing

➢Highlighting specific range of intensities Example :

▪ Enhancing features such as masses of water in the satellite

Imagery.

▪Enhancing flaws in X-ray images.

Piecewise Linear transformation functions

Intensity Level slicing

L-1

T(r)
s

0 r
L-1
Piecewise Linear transformation functions

Intensity Level slicing

L-1

s T(r)

0 r
L-1
 Piecewise Linear transformation functions

Intensity Level slicing (Example)

Piecewise Linear transformation functions

Intensity Level slicing (Example)

clear all ;
clc;
[filename, pathname] = uigetfile('*.tif');
im = imread([pathname filename]);
z=double(im);
[row,col]=size(z);
for i=1:1:row
for j=1:1:col
if((z(i,j)>142)) && (z(i,j)<250)
z(i,j)=255;
else
z(i,j)=im(i,j);
end
end
end
figure(1); %-----------Original Image-------------%
imshow(im);
figure(2); %-----------Gray Level Slicing With Background-------------%
imshow(uint8(z));
 Piecewise Linear transformation functions

Intensity Level slicing (Example)

Piecewise Linear transformation functions

Intensity Level slicing (Example)

clear all ;
clc;
[filename, pathname] = uigetfile('*.tif');
im = imread([pathname filename]);
z=double(im);
[row,col]=size(z);
for i=1:1:row
for j=1:1:col
if((z(i,j)>142)) && (z(i,j)<250)
z(i,j)=255;
else
z(i,j)=0;
end
end
end
figure(1); %-----------Original Image-------------%
imshow(im);
figure(2); %-----------Gray Level Slicing With Background-------------%
imshow(uint8(z));
 Piecewise Linear transformation functions

Bit Plane slicing (Example)

➢ Each pixels are digital number comprising of bits
➢ For a 256 level gray-scale image there are 8 bits for each pixel
➢ We can highlight the contribution of these bits to total image appearance

Example pixel value =135

1
0
0
10000111 0
0
1
1
1
Piecewise Linear transformation functions

Bit Plane slicing (Example)

An 8 bit gray scale

image

Contribution of bit
plane 8
Image Averaging

• A noisy image:

g ( x, y ) = f ( x, y ) + n ( x, y )

• Averaging K different noisy images:

M
1
g ( x, y ) =
K
 g ( x, y )
i =1
i
Image Averaging

• As K increases, the variability of the pixel

values at each location decreases.

–This means that g(x,y) approaches f(x,y) as

the number of noisy images used in the
averaging process increases.

• Registering(aligned) of the images is

necessary to avoid blurring in the output
image.
Histogram Modification
✓Histogram modification performs a function
similar to gray level mapping, but works by
considering histogram’s shape and spread.

✓Gray level histogram of an image is the

distribution of the gray levels in an image.

✓Examination of the histogram is one of the

most useful tools for image enhancement, as it
makes easy to see the modifications that may
improve an image.
✓The histogram can be modified by a mapping
function, which will stretch, shrink
(compress), or slide the histogram.

✓Histogram stretching and histogram shrinking

are forms of gray scale modification,
sometimes referred to as histogram scaling.
✓Histogram stretch
• The mapping function equation is as follows:

where: I(r,c)MAX is the largest gray level value in the image

I(r,c), I(r,c)MIN is the smallest gray level value in I(r,c) and

MAX and MIN correspond to the maximum and minimum

gray level values possible (for an 8-bit image these are 0

and 255)
• This equation will take an image and stretch
the histogram across the entire gray level
range, which has the effect of increasing the
contrast of a low contrast image

• If most of the pixel values in an image fall

within a small range, it is useful to allow a
small percentage of the pixel values to be
clipped at the low and high end of the range
(for an 8-bit image this means truncating at 0
and 255)
 Histogram Stretching

b) Histogram of image (a)

a) Low-contrast image

d) Histogram of image after stretch

c) Image (a) after histogram stretch
 Histogram Stretching with Clipping

a) Original image b) Histogram of original image c)Image after histogram stretching

with out clipping

d) Histogram of image (c) e) Image after histogram stretching with f) Histogram of image (e)
clipping 1% of the values at the high
and low ends
Histogram shrink

✓The mapping function equation is as follows:

where I(r,c)MAX is the largest gray level value in

the image I(r,c), I(r,c)MIN is the smallest gray
level value in I(r,c) and ShrinkMAX and
ShrinkMIN correspond to the maximum and
minimum desired in the compressed histogram.
• Decreases image contrast by compressing the
gray levels.

• However this method may not be useful as an

image enhancement tool, but it is used in an
image sharpening algorithm (unsharp masking)
as a part of an enhancement technique.
 Histogram Shrinking

a) Original image b) Histogram of image

c) Image after shrinking the histogram d) Histogram of image (c)

to the range [75,175]
✓ Histogram slide

• Used to make an image either darker or lighter, but

retain the relationship between gray level values

• Accomplished by simply adding or subtracting a

fixed number from all of the gray level values, as
follows:

where the OFFSET value is the amount to slide the histogram

• In this equation we assume that any values
slid past the minimum and maximum values
will be clipped to the respective minimum or
maximum
• A positive OFFSET value will increase the
overall brightness, while a negative OFFSET
will create a darker image
 Histogram Slide

a) Resultant image from sliding the b) Histogram of image (a)

histogram down by 50

c) Resultant image from sliding the

d) Histogram of image (c)
histogram up by 50
Histogram Equalization

pr ps
r Histogram s
Equalization
T (.)
dr
p s ( s ) = pr ( r )
ds
ps ( s ) = 1 T (.)
r
s = T (r ) =  pr (u )du
0

# pixels with the

j-th gray-level
k k nj
sk = T (rk ) =  pr (rj ) =  Point operation for
j =0 j =0 MxN equalizing histogram
image size for the example image

87
✓Histogram equalization

• A technique where the histogram of the resultant image is as flat

as possible.
• The theoretical basis for histogram equalization involves
probability theory, where we treat the histogram as the probability
distribution of the gray levels.
• Its function is similar to that of a histogram stretch but often
provides more visually pleasing results across a wider range of
images.
• Consists of four steps:

1. Find the running sum of the histogram values

2. Normalize the values from step (1) by dividing by the total
number of pixels
3. Multiply the values from step (2) by the maximum gray level
value and round
4. Map the gray level values to the results from step (3) using a
one-to-one correspondence
Example:
3-bits per pixel image – range is 0 to 7.
Given the following histogram:

Number of Pixels
Gray Level Value (Histogram values)
0 10
1 8
2 9
3 2
4 14
5 1
6 5
7 2
1) Create a running sum of the histogram values. This
means the first value is 10, the second is 10+8=18,
next 10+8+9=27, and so on. Here we get 10, 18, 27,
29, 43, 44, 49, 51
2) Normalize by dividing by the total number of
pixels. The total number of pixels is:
10+8+9+2+14+1+5+0 = 51 (note this is the last
number from step 1), so we get: 10/51, 18/51, 27/51,
29/51, 43/51, 44/51, 49/51, 51/51
3) Multiply these values by the maximum gray level
values, in this case 7, and then round the result to
the closest integer. After this is done we obtain: 1, 2,
4, 4, 6, 6, 7, 7
4) Map the original values to the results from step 3 by
a one-to-one correspondence. This is done as
follows:

Original Gray Histogram

Level Value Equalized Values
0 1
1 2
2 4
3 4
4 6
5 6
6 7
7 7
▪ All pixels in the original image with gray
level 0 are set to 1, values of 1 are set to 2, 2
set to 4, 3 set to 4, and so on. After the
histogram equalization values are calculated
and can be implemented efficiently with a
look-up-table (LUT), as discussed in Chapter
2
▪ We can see the original histogram and the
resulting histogram equalized histogram in
Fig. 8.2.14. Although the result is not flat, it is
closer to being flat than the original histogram
• Histogram equalization of a digital image will not
typically provide a histogram that is perfectly flat,
but it will make it as flat as possible.

• Histogram equalization may not always provide the

desired effect, since its goal is fixed – to distribute
the gray level values as evenly as possible. To allow
for interactive histogram manipulation, the ability to
specify the histogram is necessary.
Color Image Enhancement
 Enhancement of Color Images

• Gray scale transforms and histogram modification

techniques can be applied by treating a color image as three
gray images.

• Care must be taken in how this is done to avoid color shifts.

✓Histogram modification can be performed on
color images, but doing it on each color band
separately can create relative color changes.

✓The relative color can be retained by applying

the gray scale modification technique to one
of the color bands, and then using the ratios
from the original image to find the other
values.
 Histogram Equalization of Color Images

a) Original poor contrast image b) Histogram equalization based on

the red color band
 Histogram Equalization of Color Images (contd)

c) Histogram equalization based on d) Histogram equalization based on

the green color band the blue color band

Note: In this case the red band gives the best results
This will depend on the image and the desired result
✓Typically the most important color band is
selected, and this choice is very much
application-specific and will not always
provide us with the desired result.

✓Often, we really want to apply the gray

scale modification method to the image
brightness only, even with color images.
✓Histogram modification on color images can
be performed in the following ways:

• Retain the RGB ratios and perform the

modification on one band only, then use the
ratios to get the other two bands’ values, or
• Perform a color transform, such as HSL, do
the modification on the lightness (brightness
band), then do the inverse color transform.
 Histogram Matching

• Transform image such that resulting image has specified

histogram.

r Histogram z
Matching
pr F (.) pz

r
s = T (r ) =  pr (u )du
z
0
z = G −1 ( s) = G −1[T (r )] F = G −1T
G ( z ) =  pz (t )dt = s
0

7/31/2025 118
 Histogram Matching

k
sk = T (rk ) =  pr (rj )
j =0
k
vk = G( zk ) =  pz ( z j ) = sk
j =0

k = 0,1,, L − 1

7/31/2025 119
✓ Histogram Matching

• Process of defining a histogram and modifying

the histogram of the original image to match the
histogram as specified.

• Key concept is to picture the original image

being histogram equalized, and the specified
histogram being histogram equalized.
a

b
• Histogram specification consists of
following 5 steps:

1. Specify the desired histogram

2. Find the mapping table to histogram
equalize the image, Mapping Table 1,
3. Find the mapping table to histogram
equalize the values of the specified
histogram, Mapping Table 2.
• Histogram specification steps (continued)

4. Use mapping Tables 1 & 2 to find the

mapping table to map the original values.
to the histogram equalized values and then to
the specified histogram values.
5. Use the table from step (4) to map the
original values to the specified histogram
values.
EXAMPLE:
1) Specify the desired histogram:

Number of pixels
Gray Level Value in desired histogram
0 1
1 5
2 10
3 15
4 20
5 0
6 0
7 0
2) For this we will use the image and mapping table from
the previous example, where the histogram equalization
mapping table (Mapping Table 1) is given by:

Original Gray Level Value Histogram Equalized

level values-OS equalized values-HS

0 1
1 2
2 4
3 4
4 6
5 6
6 7
7 7
3) Find the histogram equalization mapping table
(Mapping Table 2) for the specified histogram:

Gray Level Histogram Equalized

Value - OS Values – HS .
0 round(1/51)*7 = 0
1 round(6/51)*7 = 1
2 round(16/51)*7 = 2
3 round(31/51)*7 = 4
4 round(51/51)*7 = 7
5 round(51/51)*7 = 7
6 round(51/51)*7 = 7
7 round(51/51)*7 = 7
4) Use Mapping Tables 1 and 2 to find the final mapping
table by mapping the values first to the histogram
equalized values and then to the specified histogram
values. (Mapping Table 2, columns switched to match Fig. 8.2.16 – slide 57)

Mapping Table 1 Mapping Table 2

O H HS OS M
0 1 0 0 1
1 2 1 1 2
2 4 2 2 3
3 4 4 3 3
4 6 7 4 4
5 6 7 5 4
6 7 7 6 4
7 7 7 7 4
5) Use the table from STEP 4 to perform the histogram
specification mapping. For this all we need are columns
O (or OS) and M:

O M
0 1
1 2
2 3
3 3
4 4
5 4
6 4
7 4

Now, all the 0’s get mapped to 1’s, the 1’s to 2’s, the 3’s
to 3’s and so on
• In practice, the desired histogram is often
specified by a continuous (possibly non-
linear) function, for example a sine or a log
function.

• To obtain the numbers for the specified

histogram the function is sampled, the values
are normalized to 1, and then multiplied by
the total number of pixels in the image
Histogram Matching Examples

Original
image

Histogram
of
original
image
Histogram Matching Examples (contd)

Original histogram

Specified histogram, exp(0.015*x)

Output image and its histogram
Histogram Matching Examples (contd)

Original histogram

Specified histogram, log(0.5*x+2)

Output image and its histogram
Image Enhancement in the
Spatial Domain
HOMOMORPHIC FILTERING
• It simultaneously normalizes the brightness
across an image and increases contrast.
• Filtering is used to remove multiplicative noise.

• Illumination and reflectance are not separable.

• Illumination and reflectance combine

multiplicatively.

• Components are made additive by taking the

logarithm of the image intensity.
• Multiplicative components of the image can be separated linearly in
the frequency domain

• To make the illumination of an image more even, the high-frequency

components are increased and low-frequency components are
decreased

• High-frequency components are assumed to represent mostly the

reflectance in the scene (the amount of light reflected off the object
in the scene)

• Low-frequency components are assumed to represent mostly the

illumination in the scene

• High-pass filtering is used to suppress low frequencies and amplify

high frequencies, in the log-intensity domain
• The illumination component tends to vary slowly across
the image.

• The reflectance tends to vary rapidly, particularly at

junctions of dissimilar objects.

• Therefore, by applying a frequency domain filter of the

form we can reduce intensity variation across the image
while highlighting detail.
Homographic filtering : PET example
Spatial filtering
Linear spatial filtering
On continuing ……
Image Smoothing Filter
Variable Weights
Gaussian Filters
Directional Smoothing
Conservative smoothing
• The contra-harmonic mean filter works
well for images containing salt OR pepper
type noise, depending on the filter order,
R:

• For negative values of R, it eliminates salt-

type noise, while for positive values of R, it
eliminates pepper-type noise
• The geometric mean filter works best
with Gaussian noise, and retains detail
information better than an arithmetic mean
filter

• It is defined as the product of the pixel

values within the window, raised to the
1/(N*N) power:
• The harmonic mean filter fails with pepper
noise, but works well for salt noise
• It is defined as follows:

• This filter also works with Gaussian noise,

retaining detail information better than the
arithmetic mean filter
• The Yp mean filter is defined as follows:

• This filter removes salt noise for negative values of P,

and pepper noise for positive values of P
 Median Filter

a) Image with added salt-and-pepper noise, b) After median filtering with a 3x3
the probability for salt = probability window,all the noise is not removed
for pepper = 0.10
 Median Filter

c) After median filtering with a 5x5 window, all the noise is

removed, but the image is blurry acquiring the “painted”
effect
Sharpening Filters
High-Boost Filter
Unsharp Masking
Examples of Unsharp masking
Image Enhancement in the Frequency Domain

Origin = 0,0
2D Discrete Fourier Transform
f(x,y) F(u,v) – not centered

Centered + Log
Centered F(u,v) Transformed F(u,v)

Origin = 0,0
 Image Enhancement in the Frequency Domain

2D Discrete Fourier Transform

Consider the following 2 left images which are pure vertical cosine of 4 cycles and a pure
vertical cosine of 32 cycles.
f(x,y) F(u,v) Notice that the Fourier Spectrum for
each image at the right contains just a
single component, represented by 2
bright spots symmetrically placed
about the center.

The dot at the center that represents

f(x,y) F(u,v) the (0,0) frequency term or average
value of the image. Images usually
have a large average value/DC
component. Due to low frequencies
Fourier Spectrum images usually
have a bright blob of components near
the center.

© 2002 R. C. Gonzalez & R. E. Woods

 Image Enhancement in the Frequency Domain

2D Discrete Fourier Transform

Consider the following 2 left images with pure cosines in pure forward diagonal and mixed
vertical+horizontal.
f(x,y) F(u,v)
One of the properties of the 2D
FT is that if you rotate the
image, the spectrum will rotate
in the same direction.

f(x,y) F(u,v)

The sum of 2 sine functions, each

in opposite (vertical and
horizontal) direction.

© 2002 R. C. Gonzalez & R. E. Woods

 Image Enhancement in the Frequency Domain

2D Discrete Fourier Transform

The center value (at the origin) of the Frequency Spectrum corresponds to the ZERO
frequency component which also referred to as the DC component in an image:
Substituting 0,0 to the origin, the Fourier transform function yields to the average/DC
component value as follows: M −1N −1
1
F(0,0) =
MN
 f (x, y)
f(x,y) F(u,v) x=0 y=0 f(x,y) F(u,v)

The center/zero frequency

component (DC component) is
removed
f(x,y) F(u,v)
(Average image) (DC Component)

© 2002 R. C. Gonzalez & R. E. Woods

 Image Enhancement in the Frequency Domain

2D Discrete Fourier Transform

f(x,y) F(u,v)
The lines in an image often
generate perpendicular lines
in the spectrum.

f(x,y) F(u,v)
The sloped lines in the
spectrum are due to the sharp
transition from the sky to the
mountain

© 2002 R. C. Gonzalez & R. E. Woods

 Image Enhancement in the Frequency Domain

Filtering in the Frequency Domain

•Some Basic Properties of the Frequency Domain:
•Frequency is directly related to the rate of change. Therefore, slowest varying
component (u=v=0) corresponds to the average intensity level of the image.
Corresponds to the origin of the Fourier Spectrum.

f(x,y)

•Higher frequencies corresponds to the faster

varying intensity level changes in the image. The
edges of objects or the other components
characterized by the abrupt changes in the
intensity level corresponds to higher frequencies.

F(u,v)

Prepared By: Dr. Hasan Demirel, PhD

© 2002 R. C. Gonzalez & R. E. Woods
 Image Enhancement in the Frequency Domain

Filtering in the Frequency Domain

•Basic Steps for Filtering in the Frequency Domain:

Prepared By: Dr. Hasan Demirel, PhD

© 2002 R. C. Gonzalez & R. E. Woods
Image Enhancement in the Frequency Domain

Filtering in the Frequency Domain

•Basic Steps for Filtering in the Frequency Domain:
1. Multiply the input image by (-1)x+y to center the transform.
2. Compute F(u,v), the DFT of the image from (1).
3. Multiply F(u,v) by a filter function H(u,v).
4. Compute the inverse DFT of the result in (3).
5. Obtain the real part of the result in (4).
6. Multiply the result in (5) by (-1)x+y .

Given the filter H(u,v) (filter transfer function) in the frequency domain, the Fourier
transform of the output image (filtered image) is given by:

G(u, v) = H (u, v)F(u, v) Step (3)

The filtered image g(x,y) is simply the inverse Fourier transform of G(u,v).

g(x, y) = −1 G(u, v) Step (4)

Image Enhancement in the Frequency Domain

Filtering in the Frequency Domain

•Basics of Low Pass Filters in the Frequency Domain:

•lowpass filter: A filter that attenuates

high frequencies while passing the low
frequencies.
•Low frequencies represent the gray-
level appearance of an image over
smooth areas.

•highpass filter: A filter that attenuates

low frequencies while passing the high
frequencies.
•High frequencies represents the details
such as edges and noise.
Image Enhancement in the Frequency Domain

Filtering in the Frequency Domain

•Consider the following filter transfer function:
0 if (u, v) = (M / 2, N / 2)
H (u, v) = 
1 otherwise
•This filter will set F(0,0) to zero and leave all the other frequency components. Such a
filter is called the notch filter, since it is constant function with a hole (notch) at the origin.
Image Enhancement in the Frequency Domain

Filtering in the Frequency Domain

• Smoothing Frequency Domain Filters: Ideal Lowpass Filters

1 if D(u, v)  D0
H (u, v) =  D(u,v) is the distance from the origin
0 if D(u, v)  D0 D0is the cutoff frequency
Image Enhancement in the Frequency Domain

Filtering in the Frequency Domain

• Smoothing Frequency Domain Filters: Butterworth Lowpass Filters
1 D(u,v) is the distance from the origin
H (u, v) = D0 is the cutoff frequency.
1+[D(u, v) / D0 ]2n
n is the order of the filter
Image Enhancement in the Frequency Domain

Filtering in the Frequency Domain

• Smoothing Frequency Domain Filters: Gaussian Lowpass Filters

− D2 (u,v)/ 2D 02 D(u,v) is the distance from the origin

H (u, v) = e D0 is the cutoff frequency.
Image Enhancement in the Frequency Domain

Filtering in the Frequency Domain

•Comparison of Ideal, Butterworth and Gaussian Lowpass Filters having the
same radii (cutoff frequency) value.

Original image Ideal LP Filtered Butterworth LP Filtered Gaussian LP Filtered

Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

• The high-frequency components are: edges and sharp transitions such as noise.

•Smoothing/blurring can be achieved by attenuating a specified range of high-

frequency components in the frequency domain.

•Given the Fourier transformed image F(u) , the filtered image G(u) can be
obtained by:
G(u, v) = H (u, v)F(u, v)
•Where H(u) is the filter transfer function.

•Smoothing can be achieved by lowpass filters. We will consider only 3 types of

lowpass filters:
•Ideal Lowpass filters,
•Butterworth Lowpass filters,
•Gaussian Lowpass filters
Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

•Ideal Lowpass Filter (ILPF): Simply cuts off all the high frequencies higher
than the specified cutoff frequency. The filter transfer function:

1 if D(u, v)  D0 D(u,v) is the distance from the origin

H (u, v) =  D0is the cutoff frequency
0 if D(u, v)  D0

Cutoff frequency
Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

•Cutoff Frequency of an Ideal Lowpass Filter: One way of specifying the
cutoff frequency is to compute circles enclosing specified amounts of total image
power.
•Calculating PT which is the total power of the transformed image:
M −1 N −1
PT=  P(u, v) u = 0, 1, ..., M − 1 , v = 0, 1, ..., N − 1
u=0 v=0

•The cutoff frequency Do can be determined by specifying

the  percent of the total power enclosed by a circle
centered at the origin. The radius of the circle is the cutoff
frequency Do.

 = 100  P(u, v) / PT  u  radius( D o) , v  radius( D o)

 u v 
Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

• Cutoff Frequency of an Ideal Lowpass Filter:

Original Image Fourier Spectrum 15 pixels ,  = 92%

80 pixels ,  = 98%

230 pixels ,
 = 99.5%
 Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

• Cutoff Frequency of an Ideal Lowpass Filter:

ILPF with cutoff=5, removed power=8%

Blurring effect is the result of LPIF
Original
Image

ILPF with cutoff=30, removed power=3.6%

cutoff=15,
r.d. power=5.4% Ringing effect is the problem of LPIF

ILPF with cutoff=230, removed power=0.5%

cutoff=80,
r.d. power=2%
Image Enhancement in the Frequency Domain
• G(u, v) = H (u, v)F(u, v)
• The corresponding process in the spatial domain with regard to the convolution
• theorem is given by:
• g(x, y) = h(x, y)  f (x, y)
• h(x,y) in the spatial domain, is the inverse Fourier transform of the filter transfer
function H(u,v).

Smoothing Frequency-Domain Filters

•Blurring and Ringing properties of ILPF:
•The blurring and ringing properties of the ILPF can be explained by the help of
the convolution theorem:
•Given the Fourier transformed input image F(u) and output image G(u) and
the filter transfer function H(u),

•The spatial filter h(x,y) has two major characteristics:

• dominant component at the origin
•Concentric circular components about center component.
Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

• Blurring and Ringing properties of ILPF:

1D Frequency Domain- H(u) 1D Spatial Domain- h(x)

inverse DFT

2D Frequency Domain- H(u,v) 2D Spatial Domain – h(x,y)

•The spatial domain filters center component is responsible for blurring.
•The circular components are responsible for the ringing artifacts.
Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

•Blurring and Ringing properties of ILPF: Lets consider the following
convolution in the spatial domain:

Central components: causes blurring

* =
circular components: causes ringing effect
Input image - f(x,y) Spatial Filter – h(x,y) filtered image – g(x,y)
Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

•Butterworth Lowpass Filter (BLPF): The transfer function of BLPF of
order n and with a specified cutoff frequency is denoted by the following filter
transfer function:
1
H (u, v) = D(u,v) is the distance from the origin
D0 is the cutoff frequency.
1+[D(u, v) / D0 ]2n n is the order of the filter
Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

• Butterworth Lowpass Filter (BLPF):
•The BLPF with order of 1 does not have any ringing artifact.
•BLPF with orders 2 or more shows increasing ringing effects as the order
increases.

D0 =5, n=1 D0 =5, n=2 D0 =5, n=3 D0 =5, n=4

2D - Spatial domain representation of the BLPF.

1D - Profile through the centre of the Spatial domain filters.

 Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

•Butterworth Lowpass Filter (BLPF):

BLPF with cutoff=5, order=2

Original Image

cutoff=15, BLPF with cutoff=30, order=2

order=2

BLPF with cutoff=230, order=2

cutoff=80,
order=2
Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

•Gaussian Lowpass Filter (GLPF): The transfer function of GLPF is given as
follows:
2
− D 2 (u,v)/ 2D 0
H (u, v) = e
D(u,v) is the distance from the origin
D0 is the cutoff frequency.
 Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

•(GLPF): The inverse Fourier transform of the Gaussian Lowpass

cutoff=5 filter is also Gaussian in the Spatial

domain.
Original
Image
•Therefore there is no ringing effect
of the GLPF. Ringing artifacts are
cutoff=15 cutoff=30 not acceptable in fields
like medical imaging. Hence use
Gaussian instead of the
ILPF/BLPF.

cutoff=230

cutoff=80
Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

•Gaussian Lowpass Filter (GLPF): Refer to the improvement in the following
example.
Image Enhancement in the Frequency Domain

Smoothing Frequency-Domain Filters

•Gaussian Lowpass Filter (GLPF): The following example shows a lady
younger. How? By using GLPF!
Image Enhancement in the Frequency Domain

Sharpening Frequency-Domain Filters

•The high-frequency components are: edges and sharp transitions such as noise.

•Sharpening can be achieved by highpass filtering process, which attenuates low

frequency components without disturbing the high-frequency information in the
frequency domain.

•The filter transfer function, Hhp(u,v), of a highpass filter is given by:

Hhp (u, v) = 1− Hlp (u, v)

•Where Hlp(u,v), is the transfer function of the corresponding lowpass filter.

•We will consider only 3 types of sharpening highpass filters :

•Ideal Highpass filters,
•Butterworth Highpass filters,
•Gaussian Highpass filters
Image Enhancement in the Frequency Domain

Sharpening Frequency-Domain Filters

•Ideal Highpass Filter (IHPF): Simply cuts off all the low frequencies lower
than the specified cutoff frequency. The filter transfer function:

0 if D(u, v)  D0 D(u,v) is the distance from the origin

H (u, v) =  D0is the cutoff frequency
1 if D(u, v)  D0

Perspective plot Image representation Cross section of

of IHPF of IHPF IHPF
Image Enhancement in the Frequency Domain

Sharpening Frequency-Domain Filters

•Ideal Highpass Filter (IHPF): The ringing artifacts occur at low cutoff
frequencies

D0 =15 D0 =30 D0 =80

Image Enhancement in the Frequency Domain

Sharpening Frequency-Domain Filters

•Butterworth Highpass Filter (BHPF): The transfer function of BHPF of
order n and with a specified cutoff frequency is given by:

1 D(u,v) is the distance from the origin

H (u, v) = D0 is the cutoff frequency.
1+[D0 / D(u, v)]2n n is the order of the filter

Perspective plot Image representation Cross section of

of BHPF of BHPF BHPF
Image Enhancement in the Frequency Domain

Sharpening Frequency-Domain Filters

•Butterworth Highpass Filter (BHPF): Smoother results are obtained in
BHPF when compared IHPF. There is almost no ringing artifacts.

D0 =15, n=2 D0 =30, n=2 D0 =80, n=2

 Sharpening Frequency-Domain Filters
•Gaussian Highpass Filter (GHPF): The transfer function of GHPF is given
by:
2 D(u,v) is the distance from the origin
−D 2 (u,v)/ 2D 0
H (u, v) = 1− e D0 is the cutoff frequency.

Perspective plot Image representation Cross section of

of BHPF of BHPF BHPF
Image Enhancement in the Frequency Domain

Sharpening Frequency-Domain Filters

•Gaussian Highpass Filter (GHPF): Smoother results are obtained in GHPF
when compared BHPF. There is absolutely no ringing artifacts.

D0 =15 D0 =30 D0 =80

Image Enhancement in the Frequency Domain

Sharpening Frequency-Domain Filters

•Comparison of Ideal, Butterworth and Gaussian High pass Filters
(D0 =30)

Ideal High pass Filter Butterworth High pass Filter Gaussian High pass Filter
Thank You