UNIT 3

IMAGE ENHANCEMENT IN
SPATIAL DOMAIN

Structure Page No.

5.1 Introduction 95
Objectives
5.2 Image Enhancement 96
5.3 Point Operations 97
5.4 Contrast Stretching 100
5.5 Clipping and Thresholding 101
5.6 Digital Negative 102
5.7 Intensity Levels Slicing 103
5.8 Bit Extraction 105
5.9 Summary 108
5.10 Solutions/Answers 108

5.1 INTRODUCTION
In this unit, we provide an overview of image enhancement techniques in
spatial domain. These techniques improve the quality of images. The
enhancement process does not increase the information content in the data.
But it increases the dynamic range of the selected features so that they can be
detected easily. Many point processing enhancement techniques are
suggested in this unit. Image enhancement is a very important topic because
of its usefulness in virtually all image processing applications.

Now, we shall list the objectives of this unit. After going through the unit,
please read this list again and make sure that you have achieved the
objectives.

Objectives
After studying this unit, you should be able to

 define image enhancement

95
Block 1 Digital Image Processing-I
 perform image enhancement in spatial domain
 perform point operations on images
 perform image enhancement using the following algorithms
 Point operations
 Contrast stretching
 Clipping and thresholding
 Digital Negative
 Intensity levels slicing
 Bit plane extraction

Let us begin the unit by discussing image enhancement.

5.2 IMAGE ENHANCEMENT

Image enhancement techniques improve the quality of image as perceived by
human observer/ machine vision system. Enhancement techniques improve
the perception of information in an image for human viewing and provide
‘better’ input for other automated image processing techniques. The main
objective is to modify attributes of an image to make it more suitable for a
given task or a specific observer. Image quality can degrade because of poor
illumination, improper acquisition device, coarse quantization noise during
acquisition process etc. The recorded images after acquisition exhibit
problems such as.

• Too dark
• Too light
• Not enough contrast
• Noise

Thus, enhancement aims to improve visual quality by ‘Cosmetic processing’. A

process of improving the visual quality of any image so that it is more suitable
for a particular application is termed as enhancement. The enhancement
process does not increase the inherent information content in the data. But, it
increases the dynamic range of the chosen features so that they can be
detected easily. Generally, humans are the ultimate judge of the improved
quality. Quality can also be objectively quantified (measured) by metrics like
mean square error (MSE).

Suitability of the enhanced image heavily depends on the application.

Enhancement is generally one of the preprocessing methods used on an
image so that it is more suitable for further processing. For example, a finger
print recognition system used for attendance recording of the employees in an
organization, uses image enhancement techniques to get best recognition
results under all circumstances. During finger print capturing, the quality of
finger print can go down because of dust, sweat, noise, [Link]
enhancement techniques make the input more suitable for further processing
so that best results can be achieved. If preprocessing is skipped and finger
print matching algorithm is applied directly on the input, the algorithm can
show a mismatch for an input which is otherwise a match.

Evaluation of image quality by human observer is a very subjective process

and is hard to standardize. An image may be good in one person’s opinion,
may not be good in another person’s opinion. But people’s view about the
96 quality of an image is very important and cannot be neglected. Generally, a set
Unit 3 Image Enhancement in Spatial Domain
of 20 (or larger number) people are asked to give their opinion about the
enhanced image and average results are taken.

Evaluation task for machine perception is much easier. In this case, a good
image is defined as one which gives best machine recognition results. But
certain amount of trial & error is generally required before a particular image
enhancement approach is selected.

Image enhancement techniques are application specific and produce a ‘better’

image.

They are broadly classified into two categories namely

i) Spatial domain methods

ii) Frequency domain methods

In spatial (time) domain methods, pixel values are manipulated directly to get
an enhanced image, whereas in frequency domain methods, firstly fourier
transform of image is taken to convert image into frequency domain. Then the
fourier transform is manipulated and the modified spectrum is transformed
back to spatial domain to view the enhanced image. Some enhancement
techniques operate on combination of these methods to get best results.

Based on what we have discussed so far, you may try the following exercises.

E1) Specify the objectives of image enhancement techniques.

E2) What are the two types of image enhancements? Define them with the
help of suitable examples.

E3) What is the importance of image enhancement in image processing?

In the following section, we shall discuss the point processing.

5.3 POINT OPERATIONS

In this unit, we shall discuss various spatial domain enhancement techniques.
The pixel values (grey values) of an image are directly manipulated. Such
operations simply take the grey value of each pixel/neighbouring pixels, map it
to a new value and move on to the next pixel. Let f ( x, y) be an input image.
An image processing operation in the spatial domain may be expressed as a
mathematical function T[f ( x , y)] applied to the image f ( x, y) to produce a new
image g( v, v). Therefore, g ( x , y)  T[f ( x , y)].

(a) (b) (c)

Fig. 1: Point and Neighbourhood Processing

97
Block 1 Digital Image Processing-I
We define the operator T applied on f ( x , y) over

i) a single pixel ( x , y), which is called ‘ point processing’, as shown in Fig.

1 (b).
ii) some neighbourhood of ( x , y), which is called ‘Neighbourhood
processing’, as shown in Fig. 1 (c).
iii) T may operate on a set of input images instead of a single image.

The principal approach defining a neighbourhood about a point ( x , y) is to use

a square or rectangular sub image are centred at ( x , y) as shown in Fig. 1.
The centre of the sub image is moved from pixel to pixel starting at the top left
corner. The operator T is applied to each location ( x , y) to yield the output
g ( u, v) at that location.

Point processing is the simplest case of spatial domain techniques where

output at ( x , y) only depends on the input intensity at the same point as shown
in Fig. 2. It is a memoryless operation. In this case, pixels of same intensity get
the same

transformation.

f ( x , y) g ( u , v)
Fig. 2: Point operations

Let us see how these transformations take place.

Example 1: Perform the transformation g1 ( v)  v  1 on the image

 2  1 0
f ( x , y)   .
0 1 2

 2  1  1  1 0  1  1 0 1
Solution: G1 ( x, y)  g1 (f ( x, y))    .
 0  1 1  1 2  1  1 2 3

***

Example 2: Perform the transformation g 2 ( v)  v 2 on the image

 2  1 0
f ( x , y)   .
0 1 2
(2) 2 (1) 2 0 2  4 1 0
Solution: G1 ( x, y)  g 2 (f ( x, y))   2  
 0 12 0 2  0 1 4 .

***
98
Unit 3 Image Enhancement in Spatial Domain
Point operations are simplest yet extremely useful and powerful image
processing tasks. Point operations are defined as

g (u , v)  T[f ( x , y)], also, s  T(r )

Where r  grey level of input image f ( x , y) , and
s  grey level of output image g ( x , y)

Thus, grey levels from the input image are modified with a function T which is
independent of image coordinates. Typical examples of point operations are
modifying image brightness, histogram processing etc.

Now, we shall discuss neighbourhood processing.

In this case, the transformation operator T is defined over the neighbourhood

of ( x , y) . This neighbourhood can be defined using a square/ rectangular/
circular sub image that are centred at ( x , y) . In this method, the spatial
characteristics around the pixel ( x , y) can be used which is not possible in
point operation.

In the Fig. 3, the operation on nine pixels around ( x , y) in the input image
results in manipulating the grey level values of ( x , y) in the output image.
Some examples of neighbourhood processing are spatial filtering, Laplatian
operator etc.

Fig. 3: Neighbourhood Processing

Point Processing
Point operations are zero memory operations where a given gray level values
of an individual pixel in the input image r  [1, L  1] is mapped into a gray
levels s  [1, L  1] of the pixels in the output image using the transformation
T() .

s  T(r )

Try the following exercises.

E4) What do you mean by point processing?

99
Block 1 Digital Image Processing-I
E5) Perform the transformation g1 ( v)  3v on the image
 2  1 0
f ( x , y)   .
 0 1 2

In the following section, we shall discuss contrast stretching.

5.4 CONTRAST STRETCHING

Contrast ‘c’ between two levels x 1 and x 2 are defined as the absolute values
of the difference, between x 1 and x 2  c ( x 1 , x 2 )  x 1  x 2 . Good contrast in
an image is important to distinguish the details else they will merge in
background. Low contrast images occur due to bad or non-uniform illumination
conditions or due to nonlinearity of image acquisition devices. Fig. 4 shows an
example of low contrast image where the details are lost in background.

Fig. 4: A low contrast image

Value remapping by contrast stretching is a process that expands the range of

levels so that all the levels contribute to the image. The main idea is to reduce
the low level gray values, and to increase the mid range gray values, so that
an artificial contrast is created between the two sets of gray values. The
contrast between gray values is thus stretched from both sides.
The transformation is defined mathematically as follows:

 r 0ra

s  ( r  a )  s a arb
 (r  b)  s brL
 b

Where , ,  are slopes of different regions as shown in Fig. 5 (a). The slopes
determine the amount of contrast stretching (or diminishing). If the slope is
greater than one, then corresponding grey levels are stretched because the
original grey levels are mapped onto a larger range of grey values. A slope
smaller than one means the contrast is diminished. A slope of one indicates no
contrast alteration. The parameters a and b are user defined. Fig. 6 shows an
example of image enhancement using contrast stretching.

100
Unit 3 Image Enhancement in Spatial Domain

(a) Contrast stretching (b) Clipping (c) Thresholding

Fig. 5

(a) low contrast image (b) enhanced image

Fig. 6: Effect of Contrast Stretching

Try the following exercise.

E6) What is contrast stretching?

E7) Differentiate between low contrast image and enhanced image.

In the following section, we shall discuss clipping and thresholding.

5.5 CLIPPING AND THRESHOLDING

Clipping and thresholding are two special cases of contrast stretching.
a) Clipping
If     0 , this is called ‘clipping’ (windowing) as shown in, Fig. 5 (b). This
operation stretches the contrast to its maximum in a limited range (‘window’) of
original grey level values {a ,......, b} . All the grey levels outside this window
are either mapped to zero or the maximum values. Thus, clipping allows us to
focus all the available contrast onto the required range of grey level values.
This is especially useful when viewing medical images such as CT images. It
also helps in viewing under or over exposed images.

b) Thresholding
If     0 and a  b  t , this remapping is called ‘thresholding’ as shown in
Fig. 5 (c). This operation ‘binarizes’ (only two values present) the image. A
suitable threshold value ‘t’ is chosen, all the gray levels smaller than ‘t’ are
mapped to zero whereas all the grey levels greater than or equal to ‘t’ are
mapped to maximum value. This is a very useful operation in image
processing applications. Binarization is generally done before segmentation or

101
Block 1 Digital Image Processing-I
region extraction. Fig. 7 shows an example of thresholding with t  110 . Note
that input image is a gray scale image and output image is binary.

(a) original image (b) image after thresholding

Fig. 7: Effect of Thresholding

Example 3: Perform thresholding on image segment f ( x , y) with t  128 .

Solution:
0 10 50 100 0 0 0 0
5 95 150 200
 0 0 255 255
t 128

110 150 190 210 0 255 255 255

175 210 255 100 255 255 255 255

***
Try the following exercises.

E8) What are the two special cases of contrast stretching?

E9) Why thresholding operation results in binary output?

In the following section, we shall discuss digital negatives.

5.6 DIGITAL NEGATIVES

The digital negative of an image with the intensity levels in the range [0, L  1]
is obtained by using negative transformation shown in Fig. 8, which is given by
the expression

s  (L  1)  r

102 Fig. 8
Unit 3 Image Enhancement in Spatial Domain

In this transformation, highest grey level is mapped to lowest and vice versa.
For an8-bit image, the transformation is

s  255  r
If r  20, (dark pixel) s  255  20  235 (bright pixel)
If r  125 , s  255  125  130

Note that middle grey level has not changed much whereas dark grey level
has become bright. This transformation is generally used to enhance the white
details embedded in the dark regions of an image where black is dominant. It
is useful in displaying medical [Link]. shows an example. Even though
the visual contents of both images are same, it is much easier to analyse
image in negative form.

(a) original image (b) digital negative of fig (a)

Fig. 9: Effect of digital negative

Example 4: Find the image negative transformation on an image f given as

0 10 50 100
5 95 150 200
f ( x , y) 
110 150 190 210
175 210 255 100

Solution: The negative of the image is given as

255 245 205 155

250 160 105 55
g ( x , y) 
145 105 165 45
80 45 0 155

***
Try the following exercise.

1 2 10
E10)

Find the negative transformation of the image f ( x , y)  3 4 0 .

 
1 5 6 

103
Block 1 Digital Image Processing-I
In the following section, we shall discuss about intensity level slicing.

5.7 INTENSITY LEVEL SLICING

A variant of thresholding is intensity level slicing or double thresholding. It is
used to highlight a range of intensity values of interest in an image. In this
operation, all the grey levels in a window {a ,....., b} are set to maximum grey
level and all the grey levels outside this range are set to zero as shown in
Fig. 10.

M arb
s
r otherwise

Fig. 10: Transfer Function of Intensity Level Slicing

Intensity level slicing is a binarization operation as the resulting image has

only two grey level values (0 and M). Mostly this technique is used to remove
unwanted elements (clutter) from an image so that the useful information
becomes prominent.

Applications are enhancing certain features which are based on grey level
values, such as mass of water in satellite images, cancer cells in a monogram
images etc.

Sometimes, we wish to retain the original image as ‘background’ and highlight

the grey levels in a window, following transformation can be used

M arb
s
r otherwise

Fig. 11: Transfer Function of Intensity Level Slicing

Example 5: Perform intensity level slicing on f ( x , y) as given in Example 4,

based on transfer function shown below

104
Unit 3 Image Enhancement in Spatial Domain

Solution: In this case, grey levels less than 90 and greater than 160 are
rounded to zero and grey levels between this range are set to 220. This
transformation produces a ‘binary’ image. In this process, all the background
is lost.
0 10 50 100 0 0 0 220
5 95 150 200 T ( r ) 0 220 220 0
s 
110 150 190 210 220 220 0 0
175 210 255 100 0 0 0 220
f ( x , y) g ( x , y)
***
Try the following exercise.

E11) Perform intensity level slicing on f ( x , y) as given in Example 4, based

on transformation shown below:

In the following section we shall be discussing another sliding called it

extraction.

5.8 BIT EXTRACTION (BIT PLANE SLICING)

Generally, a grey scale image consists of 8 bits for pixel representation. Thus,
for overall appearance of the image, there is contribution by each of the 8 bits.
For a pixel P at coordinates ( x, y), contribution from each bit plane is
f ( x , y)  k 1 2 7  k 2 2 6  k 3 2 5  k 4 2 4  k 3 2 3  k 6 2 2  k 8 2 0
where k1 ,...., k 8 are either 0 or 1. Images can be assumed to be composed of
8 one bit planes with plane 1 containing the lowest order bits of all pixels in the
image and plane 8 containing all highest order bits (fig 12). Fig 13 (a) to (h)
show various bit planes of the image in fig 13 (i). Observe that four highest
order bit planes have most of the visually significant data. The lower order
planes contribute to more subtle intensity level details in the image. For

105
Block 1 Digital Image Processing-I
example, if pixel value is 194 (11000010 in binary form) then values of
k1 , k 2 ,....., k 8 are 1, 1, 0, 0, 0, 0, 1, 0.
As the contribution of higher order planes are much more in the image,
reconstruction by only these planes results to an image very close to the
original image to a great extent. To reconstruct the image using only 8th and
7th plane only is done by multiplying bit plane 8 by 128, bit plane 7 by 64 and
then adding the two planes.

Fig. 12: Bit Plane Slicing

(a) LSB plane (b) 2nd plane (c) 3rdpalne

th
(d) 4 plane (e) 5th plane (f) 6th plane

th
(g) 7 plane (h) msb plane (i) original image

Fig. 13: Bit Plane Slicing Output

Example 6: Find MSB and LSB planes for the given image

106
Unit 3 Image Enhancement in Spatial Domain
255 138 30
65 12 201
183 111 85

Solution: Represent all grey level values in binary format.

LSB MSB
255  1 1 1 1 1 1 1 1
138  0 1 0 1 0 0 0 1
30  1 1 1 1 1 0 0 0
65  1 0 0 0 0 0 1 0
12  0 0 1 1 0 0 0 0
201  1 0 0 1 0 0 1 1
183  1 1 1 0 1 1 0 1
111  1 1 1 1 0 1 1 0
85  1 0 1 0 1 0 1 0
MSB plane consists of MSB’s of all grey levels as shown in Fig. 14(a) similarly
LSB plane consists of LSB’s of all grey level values (Fig. 14(b)).

1 1 0 1 0 1
0 0 1 1 0 1
1 0 0 1 1 1
MSB plane LSB plane
(a) (b)
Fig. 14
***
Example 7: Compute various bit planes of the following 8-bit image.

0 10 50 100
50 95 150 200
110 150 190 210
175 210 255 110

Solution: To decompose grey level value 100 into 8-bit plane, convert 100
into binary  01100100 .
Then bit for 8th plane is 0,7th plane is 1, 6th plane is 1 and so on. Similarly,
binary representation of 50 is 00110010. Then bit for 8th plane 0, 7th plane is
0, 6th plane is 1 and so on.
Thus, to decompose image grey levels into various bit planes, two steps are
followed:
a) Convert the grey level value into binary.
b) Allocate bits into various planes starting from MSB.
Various bit planes for the given image are:

107
Block 1 Digital Image Processing-I
th th
8 plane ( MSB) 7 plane 6 th plane
0 0 0 0 0 0 0 1 0 0 1 1
0 0 1 1 0 1 0 1 1 0 0 0
0 1 1 1 1 0 0 1 1 0 1 0
1 1 1 0 0 1 1 1 1 0 1 1

5 th plane 4 th plane 3rd plane

0 0 0 0 0 0 0 1 0 0 1 1
0 0 1 1 0 1 0 1 1 0 0 0
0 1 1 1 1 0 0 1 1 0 1 0
1 1 1 0 0 1 1 1 1 0 1 1

2 nd plane 1st plane

0 1 1 0 0 0 0 0
1 1 1 0 0 1 0 0
1 1 1 1 0 0 0 0
1 1 1 1 1 0 1 0

***
Try the following exercises.

E12) From all bit plane of the Example 4, generate the original image.

E13) Generate the image using only 8th, 7th plane and 6th plane.

E14) What is meant by bit plane slicing?

Now, we shall summarise the unit.

5.9 SUMMARY
In this unit, we have discussed the following points.

1. Stated Image enhancement in spatial domain.

2. Explained point processing and neighbourhood processing
3. Explained various point operations such as Contrast stretching, Clipping
and thresholding, Digital Negative, Intensity levels slicing, Bit plane
extraction.
4. Used special cases of Contrast stretching: Clipping and thresholding,
which are very import preprocessing steps in image processing
algorithms.

5.10 SOLUTIONS/ANSWERS
E1) Enhancement techniques improve the perception of information in an
image for human viewing and provide ‘better’ input for other automated
image processing techniques. The main objective is to modify
108
Unit 3 Image Enhancement in Spatial Domain
attributes of an image to make it more suitable for a given task or a
specific observer.

E2) Image enhancement can be broadly classified into two categories

namely
i. Spatial domain methods
ii. Frequency domain methods

E3) Image enhancement techniques improve the quality of image as

received by human observer/ machine vision system. The
enhancement process does not increase the information content in the
data. But it increases the dynamic range of the selected features so
that they can be detected easily. Many point processing enhancement
techniques are suggested in this unit. Image enhancement is a very
important topic because of its usefulness in virtually all image
processing applications.

E4) Point operations are zero memory operations where a given gray level
values of an individual pixel in the input image r  [1, L  1] is mapped
into a gray levels s  [1, L  1] of the pixels in the output image using
s  T(r ) .

E5) G 1 ( x, y)  g 2 (f ( x, y))
  6  3 0

0 3 6
E6) Is the transformation for contrast stretching. It is used to improve the
contrast of low contrast images.

r 0ra

s  (r  a )  s a a  r  b
  ( r  b)  s b  r  1
 b

E7) You may like to differentiate low contrast image and enhance image
yourself.

E8) Clipping and thresholding are two special cases of contrast stretching.

E9) Thresholding results in binary output because the output can be either
0 or 255. Multiple input gray level values are mapped to only two
output vales.
254 253 245
E10) The negative transformation is 252 251 255 .
 
254 250 249

E11)
0 10 50 100 0 10 50 220
5 95 150 200 T ( r ) 50 220 220 200
s 
110 150 190 210 220 220 190 210
175 210 255 100 175 210 255 220
f ( x , y) g ( x , y)

109
Block 1 Digital Image Processing-I
In this case, grey levels in the window {90, 160} are set to 220 (the
circled pixels in g ( x, y)) , but other grey levels are not changed. This
transformation does not produce a binary image.

E12) To generate original image back from the bit planes we need to
multiply 128 to 8th bit, 64 to 7th bit, 32 to 6th bit, 16 to 5th bit, 8 to 4th
bit, 4 to 3rd bit, 2 to 2nd bit and 1 to 1st bit for each pixel and add all of
it.

Thus to get 100 = 128 × 0 + 64 × 1 + 32 × 1 + 16 × 0 + 8 × 0 + 4 × 1 +

2× 0 + 1 × 0.

Similarly 50 = 128 × 0 + 64 × 1 + 32 × 1 + 16 × 0 + 8 × 0 + 4 × 0 +2 × 0
+ 1 × 0 and so on the result is original image is exactly reconstructed
back as shown below.

0 10 50 100
50 95 150 200
110 150 190 210
175 210 255 110

E13) As we have to consider only 8th 7th and 6th plane, only these plane
values are added, remaining values are not considered.

g (0,0)  128  0  64  0  32  0  0
g (0,1)  128  0  64  0  32  0  0
g (0,2)  128  0  64  0  32  1  32
g (0,3)  128  0  64  1  32  1  96
g (1,0)  128  0  64  0  32  1  32

And so on. The resulting image is Fig. 16 (a). There is very small error
between the reconstructed image g ( x , y) and original image f ( x , y)
(Fig. 13 (i)). To find the error, we can find e( x, y)  f ( x , y)  g ( x , y)
which is shown in Fig. 16(b). Thus, with only 8th, 7th and 6th plane, the
image is very close to original image. The quality keeps on increasing
as more and more bit planes are added.

0 0 32 96 0 10 18 4
32 64 128 196 18 31 22 8
96 128 160 192 14 22 30 18
128 192 224 96 47 18 31 14
g ( x , y) e( x , y )

110
Unit 3 Image Enhancement in Spatial Domain

th th th
(a) image reconstructed with 7 (b) error image 6 and 5 plane
Fig. 16
E14) A grey scale image consists of 8 bits for pixel representation. Thus, for
overall appearance of the image, there is contribution by each bit of
each pixel. For a pixel P at coordinates ( x , y) , contribution from each
bit plane is

f (x, y)  k1 2 7  k 2 2 6  k 3 25  k 4 2 4  k 3 23  k 6 2 2  k 8 20

111