Image Enhancement in The Spatial Domain: 4border Code

4border code=
Image Enhancement in the Spatial Domain

Lec 03: Intensity Transformations
Dr. Ch Vit Nht Anh

nhat-anh.che@{hcmut.edu.vn,hotmail.com}
Department of Telecommunications Engineering
Faculty of Electrical and Electronic Engineerings
Ho Chi Minh City, University of Technology
HK I, 2015 - 2016
Dr. Che Viet Nhat Anh
Digital Image & Speech Processing
DI&SP - 405017
1 / 91
Outline
1 Introduction
2 Histogram
3 Some Basic Intensity Transformation Functions
4 Histogram Processing
DI&SP - 405017
2 / 91
Learning Outcomes
After learning this chapter, you will be able to:
Illustrate the effect of power-law & logarithmic histogram transforms
Describe the basis of the histogram as a Probability Density
Function (PDF)
Distinguish between the brightness and the contrast of an image
Explain the relationship between dynamic range and contrast
Explain the effects of histogram stretch, histogram equalization,
histogram matching, histogram modification & distinguish between
them
DI&SP - 405017
3 / 91
"It makes all the difference whether one sees

darkness through the light or brightness through the
shadows
David Lindsay (Scottish Novelist)
DI&SP - 405017
4 / 91
Outline
1 Introduction
- Motivation
- Type of operations
- Spatial domain
DI&SP - 405017
5 / 91
DI&SP - 405017
6 / 91
Introduction
Motivation
Which one looks better?
Introduction
Motivation
DI&SP - 405017
7 / 91
DI&SP - 405017
8 / 91
Introduction
Motivation
Introduction
Motivation
9 / 91
DI&SP - 405017
Introduction
Motivation
Close-by pixel values are difficult to distinguish.

Far apart pixel values are easy to distinguish.
Low contrast: image values concentrated near a narrow range
(mostly dark, or mostly bright, or mostly medium values).

Contrast enhancement change the image value distribution to
cover a wide range.

Contrast of an image can be revealed by its histogram.
DI&SP - 405017
10 / 91
Introduction
Type of operations
DI&SP - 405017
11 / 91
Introduction
Type of operations
Operation
Point
Local
Global
Characterization
The output value at a specific coordinate is dependent only on the
input value at that same coordinate
The output value at a specific coordinate is dependent on the input
values in the neighborhood of that
same coordinate.
The output value at a specific coordinate is dependent on all the values
in the input image.
Generic Complexity/Pixel
Constant

Image size = ; Neighborhood size =

.
The complexity is specified in operations per pixel.
DI&SP - 405017
12 / 91
Introduction
Type of operations
Domains
Spatial Domain Processing
DI&SP - 405017
13 / 91
Introduction
Type of operations
Spatial domain: Image plane itself, directly process the intensity
values of the image plane.

Point Operation (Intensity Transformations Function)
Gamma correction, window-center correction, histogram
equalization,...
Spatial filters (Neighborhood Operation)
Correlation
Convolution (Filtering): mean, Gaussian, median, etc.
Image gradients,...
Transform domain (Frequency domain, Wavelet domain) Process
the transform coefficients, not directly process the intensity values

of the image plane.
DI&SP - 405017
14 / 91
Introduction
Type of operations
Point processing
Spatial processing
Contrast stretching:
Neighborhood
cessing:
Gray level transform
pro-
Frequency
domain
processing
Linear filtering:
Low pass
Clipping
Averaging
Intensity Level slicing
Directional
High-pass
nonlinear
filtering
smoothing
Median filtering
Homomorphic
filtering
Unsharp
masking
Histogram techniques:
Equalization
Histogram specification
(matching)
15 / 91
DI&SP - 405017
Introduction
Spatial domain
Spatial domain refers to the aggregated of pixels composing an
image.
Spatial domain methods are procedures that operate directly on
these pixels. Spatial domain processes will be denoted by the
expression:
,
{ , }
(1.1)
: an operator defined over some neighborhood of ( , ), maybe
linear or nonlinear.
(0,0)
(0,0)
DI&SP - 405017
16 / 91
Histogram
Histogram of a monochrome image with gray levels in the range
, :
A discrete function, ,
, , . . . , .
: the th gray level.
the number of pixels in the image having gray level .
Normalized histogram: An estimate of the probability of
occurrence of gray level
, , , . . . , .
The sum of all components of a normalized histogram is equal to 1.
DI&SP - 405017
17 / 91
Example
Sketch the histogram of the following image:
0
5
6
0
6
2
0
6
2
7
1
6
1
0
6
0
7
2
5
1
5
1
5
6
0
4
6
1
4
2
3
3
0
4
1
5
3
6
7
7
2
5
4
3
2
1
0
0 1 2 3 4 5 6 7
DI&SP - 405017
18 / 91
Example
Sketch the normalized histogram of the following image:
6
0
6
2
0
0
0.2
1
0.16
6
2
7
1
6
2
0.12
1
0
6
0
7
3
0
2
5
1
5
1
4
0.04
5
6
0
4
6
0.28
5
0.12
6
0.28
7
0.08
0.2
0.16
0.12
0.08
0.04
0
0 1 2 3 4 5 6 7
DI&SP - 405017
19 / 91
Histogram
The following two images have the same histograms:
The histogram shows the number of pixels that have each pixel
value, but it does not record where those pixels are located in the
image. Thus, spatial information is discarded.
The histogram is unique for a particular image, but different images
could have the same histogram.

DI&SP - 405017
20 / 91
Histogram
The information inherent in histograms also is quite useful in other
image processing applications:

Image compression
Segmentation
Histograms are simple to calculate in software & also lend
themselves to economic hardware implementations, thus making

them a popular tool for real-time image processing.
The shape of the histogram of an image does provide useful
information about the possibility for contrast enhancement.

The role of histogram processing in image enhancement:
Dark image
Light image
Low contrast
High contrast
DI&SP - 405017
21 / 91
Histogram
Dark image

DI&SP - 405017
22 / 91
Histogram
Light image

DI&SP - 405017
23 / 91
Histogram
Low contrast

DI&SP - 405017
24 / 91
Histogram
High contrast

DI&SP - 405017
25 / 91
Histogram
Dynamic range:
The range of pixel values, defined as the difference between the
maximum ( ) the minimum ( ) pixel values found in the
image, ignoring any obvious outliers.
It can be expressed either as the difference in pixel values or dB:
Dynamic range of image


DI&SP - 405017
(2.1)
26 / 91
Outline
3 Some Basic Intensity Transformation Functions
- Point operations
- Identity
- Image Negatives
- Gamma Transformations
- Log Transformations
- Piecewise-Linear Transformations
- Bit Plane Slicing
27 / 91
DI&SP - 405017
Some Basic Intensity Transformation Functions

Point operations
depends only on the value of at , . becomes a gray-level
(also called an intensity or mapping) transformation function of the

form:
, { , }
(3.1)
Point operations are zero memory operations.
(0,0)
(0,0)
DI&SP - 405017
28 / 91

Point operations
DI&SP - 405017
29 / 91

Identity
No change on pixel values,
DI&SP - 405017
30 / 91

Image Negatives
Contrast reversal: negative,
It produces the equivalent of a photographic negatives.
DI&SP - 405017
31 / 91

Gamma Transformations
Power-Law transformations:
or
, , > 0.
Values of such that < < map a narrow range of dark input
pixels vales into a wider range of output values, with opposite being
true for higher values of input levels (brighten the intensity).
Values > 1 have the opposite behavior to above (darken the
intensity).

gives the identity transformation.
Gamma correction is an application.
DI&SP - 405017
32 / 91

for various values of . All curves were

scaled to fit in the range shown.
Plots of the equation
DI&SP - 405017
33 / 91

DI&SP - 405017
34 / 91

DI&SP - 405017
35 / 91

DI&SP - 405017
36 / 91

DI&SP - 405017
37 / 91

DI&SP - 405017
38 / 91

Log Transformations
Maps a narrow range of low intensity values in input to a wider
output range.
The opposite is true for high intensity input values.
Compresses the dynamic range of images with large variations in
pixel values. (The dynamic range of the image data may be very
large -> Dynamic range can be compresses via the logarithmic
transformation).
DI&SP - 405017
39 / 91

Log Transformations
DI&SP - 405017
40 / 91

Log Transformations
DI&SP - 405017
41 / 91

Piecewise-Linear Transformations: Contrast Stretching
Expands the range of intensity levels in an image so that it spans
the full intensity range of the recording medium or display device.
DI&SP - 405017
42 / 91

Low contrast images can result from poor illumination.

Reduce intensity: (, )( , ), (
,
)( , ).
Increase intensity: The slope of the transformation is chosen greater
than unity in the region of stretch.

<

<

DI&SP - 405017
(3.2)
43 / 91

DI&SP - 405017
44 / 91

Piecewise-Linear Transformations: Contrast Compressing
DI&SP - 405017
45 / 91

Piecewise-Linear Transformations: Intensity-level Slicing
Highlighting a specific range of intensities in an image.

Two main approaches:
Display in one value (e.g. white) all the values in the range of interest
and all others in another value (e.g black all other intensities)
binary image.
Brighten or darkens the desired range of intensity levels in the image
unchanged.
DI&SP - 405017
46 / 91

This transformation
highlights intensity range

, and reduces all others
intensities to a lower level.
This transformation
highlights intensity range

, and preserves all others
intensities to a lower level.
DI&SP - 405017
47 / 91

DI&SP - 405017
48 / 91

Bit Plane Slicing
Pixels are digital numbers composed of bits.

Suppose that each pixel in an image is represented by 8 bits

. Imagine that the image is composed of eight 1-bit
planes, ranging from bit-plane 0, the Least Significant Bit
(Lowest-order bit, ) to bit-plane 7, the Most Significant Bit
(Highest-order bit, .
Each bit plane is a binary image
49 / 91
DI&SP - 405017

Bit Plane Slicing
Highlight the contribution made to total image appearance by
specific bits
Higher-order bits usually contain most of the significant visual
information.
Lower-order bits contain subtle details.
Useful for compression.
We have to use bit get and bit set to extract 8 images:

136
24
33
11
67
17
1
7
3
0
0
1
1
1
1
1
1
1
0
0
0
1
1
0
0
1
1
0
0
0
0
0
0
0
1
0
1
1
0
1
0
0
0
0
0
0
1
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
1
0
0
0
0
0
0
0
0
DI&SP - 405017
50 / 91

Bit Plane Slicing
DI&SP - 405017
51 / 91

Bit Plane Slicing
Reconstruction is obtained by:
DI&SP - 405017
52 / 91
Outline
4 Histogram Processing
- Histogram Equalization
- Histogram matching or Histogram specification
- Local Histogram Processing
- Using Histogram Statistics for Image Enhancement
DI&SP - 405017
53 / 91
Histogram Processing
Histogram Equalization
Consider for a moment continuous intensity values, & let the
variable represent the gray levels of the image to be enhanced,

that is , , with = 0 representing black, &
representing white.
For any satisfying the aforementioned conditions, we focus
attention on transformations of the form:
6 6
(4.1)
Let , denote the Probability Density Function (PDF) of
random variables and .


DI&SP - 405017
54 / 91
Transforms an image with an arbitrary histogram to one with a flat
histogram.
The transformation function satisfies the following conditions:
is a monotonically increasing function in the interval
6 6 .
6 6 for 6 6 .
is continuous and differentiable.
DI&SP - 405017
55 / 91



DI&SP - 405017
56 / 91
Continuous case:
(4.2)
: a dummy variable of integration.

is high, has a steep slope, will be wide, causing to
be low to keep .
is low, has a shallow slope, will be narrow, causing
to be high to keep .
DI&SP - 405017
57 / 91
Example
Suppose that the (continuous) intensity values in an image have the
Probability Density Function (PDF)
for 6 6

otherwise
Find the transformation function for equalizing the image histogram.

DI&SP - 405017
58 / 91
Example
The histograms of two images are illustrated below. Sketch a
transformation function for each image that will make the image has a
better contrast. Use the axis provided below to sketch your
transformation functions.

20
128
100
200
100
200
20
128
DI&SP - 405017
59 / 91
Discrete values: For
, , . . . ,
(4.3)
(4.4)
The histogram equalization process 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 result from step 3 using one-to-one
correspondence.
DI&SP - 405017
60 / 91
Example
Suppose that a 3-bit image ( = 8 intensity levels) of size 64 64 pixels
( = 4096) has the intensity distribution shown in following table.
0
790
1
1023
2
850
3
656
4
329
5
245
6
122
7
81
Sketch the original histogram.
Get & sketch the histogram equalization transformation function.
Give the for each . Sketch the equalized histogram.
DI&SP - 405017
61 / 91
Example
1
Sketch the original histogram.
0
0.193
1
0.25
2
0.208
3
0.16
4
0.08
5
0.06
6
0.03
7
0.02

0.25
0.2
0.15
0.1
0.05
0 1 2 3 4 5 6 7
DI&SP - 405017
62 / 91
Example
2
Get & sketch the histogram equalization transformation function.
790
1813
2663
3319
3648
3893
4015
4096
1.33
3.08
4.55
5.67
6.03
6.65
6.86

7
5.67
4.55
7
6
5
3.08
1.33
1 2 3 4 5 6 7
1 2 3 4 5 6 7
DI&SP - 405017
63 / 91
Example
3
Give the for each . Sketch the equalized histogram.
0
0
0
1
790
0.19
2
0
0
3
1023
0.25
4
0
0
5
850
0.20
6
985
0.24
7
448
0.12

0.25
0.2
0.15
0.1
0.05
0 1 2 3 4 5 6 7
DI&SP - 405017
64 / 91
DI&SP - 405017
65 / 91
DI&SP - 405017
66 / 91


DI&SP
- 405017
67 / 91


DI&SP
- 405017
68 / 91
Histogram
Transformation function for histogram equalization:
DI&SP - 405017
69 / 91
The above procedure is for grayscale images

For color images:
Apply separately the procedure to each of the three channels
will yield in a dramatic change in the color balance
NOT RECOMMENDED unless you have a good reason for doing
that.
Recommended method
Convert the image into another color space such as the Lab color
space OR HSL/HSV color space.
Apply the histogram equalization on the Luminance channel
Convert back into the RGB color space.
DI&SP - 405017
70 / 91
Histogram equalization
Can contrast stretching achieve similar result as histogram
equalization?
If it can, why histogram equalization then?
Histogram equalization (HE) results are similar to contrast stretching
but offer the advantage of full automation, since HE automatically

determines a transformation function to produce a new image with a
uniform histogram.
Histogram equalization method:
Only generates one result: an image with approximately uniform
histogram (without any flexibility).
Enhancement may not be achieved as desired.
DI&SP - 405017
71 / 91
Histogram matching or Histogram specification
What if the desired histogram is not flat?

Histogram specification: Transform an image according to a
specified gray-level histogram

Let and denote the continuous probability density
functions of the variables and .

is the specified probability density function or specify
particular histogram shapes capable of highlighting certain
gray-level ranges.
Obtain the transformation function for transformation of to .
Histogram
Matching
DI&SP - 405017
72 / 91
Let be the random variable with the probability:
(4.5)
Define a random variable with the probability
(4.6)
The inverse of the above transform is
Histogram
Equalization

(4.7)
Histogram
Equalization
DI&SP - 405017
73 / 91
Example
Assuming continuous intensity values, suppose that an image has the
intensity PDF:
for 6 6

otherwise
Find the transformation function that will produce an image whose
intensity PDF is:
for 6 6

otherwise
DI&SP - 405017
74 / 91
Example
1
Find the histogram equalization transformation for the input image

Find the histogram equalization transformation for the specified

histogram

The transformation function

] /
[
] / [

DI&SP - 405017
75 / 91
Discrete Cases:
Obtain from the input image then obtain the values of ,
round the value to the integer range , :
(4.8)
Use the specified PDF and obtain the transformation function ,
round the value to the integer range , :

(4.9)
Find the smallest value of so that is closest to :
(4.10)
DI&SP - 405017
76 / 91
Example
Suppose that a 3-bit image ( = 8) of size 64 64 pixels ( = 4096)
has the intensity distribution shown in the following table (on the left).
Get the histogram transformation function and make the output image
with the specified histogram, listed in the table at the bottom.
0
790
1
1023
0
0
2
850
1
0
2
0
3
656
3
0.15
4
329
4
0.2
5
245
5
0.3
6
122
6
0.2
7
81
7
0.15
77 / 91
DI&SP - 405017
Example
1
Obtain the scaled histogram-equalized values:
790
1813
2663
3319
3648
3893
4015
4096

1.33
1
3.08
3
4.55
5
5.67
6
6.03
6
6.65
7
6.86
7
7
7
Compute all the values of the transformation function

0.15
0.35
0.65
0.85
0
0
0
0
0
0
1.05
1
2.45
2
3.55
5
5.95
6
7
7

DI&SP - 405017
78 / 91
Example
3
Relate the two mapping above to build a lookup table for the over
all mapping. Specifically, for each input level , find an output level
so that best matches .

0
0
0
1
0
0
0
1
0
1
3
0
2
5
0
2
0
0
3
790
0.19
3
6
1
4
6
2
4
1023
0.25
0.3
0.2
0.15
5
7
5
6
7
6
5
850
0.21
7
7
7
6
985
0.24
7
448
0.11

0.25
0.19
0.11
0 1 2 3 4 5 6 7 Digital
0 1 2 3 DI&SP
4 5 6- 405017
7
Image & Speech Processing
DI&SP - 405017
79 / 91
80 / 91
DI&SP - 405017
81 / 91
DI&SP - 405017
82 / 91
Issues with Histogram specification/matching:

No rule for specifying an optimal histogram.
Each given enhancement task needs to be analyzed on a case-by-case
basis.
Histogram specification is somehow a trial-and-error process.
DI&SP - 405017
83 / 91
Local Histogram Processing
The histogram processing methods mentioned up to now are global
transformation where:
Function is designed according to the gray-level distribution over an
entire image
Global transformation methods may not be suitable for enhancing
details over small areas
Where number of pixels in these small areas may have negligible
influence on designing the global transformation function
To enhance details over small areas in an image
Procedure:
Define a neighborhood (e.g. )
Move it from pixel to pixel.
For every pixel:
Histogram computed for the neighborhood.
Transfer function computed for Histogram sHistogram specification
or Histogram specification.
Applied on Centre Pixel.
DI&SP - 405017
84 / 91
Local Histogram Processing
DI&SP - 405017
85 / 91
Using Histogram Statistics for Image Enhancement
Global statistics:
The mean (average) intensity is given by

,

(4.11)
The variance of the intensities is given by

,

(4.12)
The th moment of the intensity variable is

(4.13)
DI&SP - 405017
86 / 91
Mean gives the average brightness of the image.

Variance
and its square root the standard deviation gives the
deviation of intensities on average from the mean value (average

contrast).
.
.
DI&SP - 405017
87 / 91
Local Statistics: : a neighborhood (subimage) of specific size
centered at , :
Local average intensity:
(4.14)

(4.15)
Local variance:
DI&SP - 405017
88 / 91
The statistical parameters can
be used in various ways.

Enhance the background
filament.
Enhance details in dark areas
while leaving light area

unchanged.
Define rules to chose the
candidate pixels that need to

be enhanced.
DI&SP - 405017
89 / 91
A pixel at point , is considered if:

6 , where is a positive constant less than 1.0, and
is global mean.
6
, where is the global standard deviation and
is a
positive constant.
Also need to put a lower limit on standard deviation to avoid
distorting areas which dont have details, i.e., 6 , with
<
.
A pixel that meets all above conditions is processed simply by
multiplying it by a specified constant, , to increase or decrease the

value of its gray level relative to the rest of the image.
The values of pixels that do not meet the enhancement conditions
are left unchanged.
{
, if < and < <

,
,
otherwise
(4.16)
DI&SP - 405017
90 / 91
DI&SP - 405017
91 / 91

Image Enhancement in The Spatial Domain: 4border Code

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Image Enhancement in The Spatial Domain: 4border Code

Hochgeladen von

Copyright:

Verfügbare Formate

4border code=

Image Enhancement in the Spatial Domain

Dr. Ch Vit Nht Anh

Digital Image & Speech Processing

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

"It makes all the difference whether one sees

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Which one looks better?

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Which one looks better?

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Which one looks better?

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Which one looks better?

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Close-by pixel values are difficult to distinguish.

(mostly dark, or mostly bright, or mostly medium values).

cover a wide range.

Digital Image & Speech Processing

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Image size = ; Neighborhood size =

Digital Image & Speech Processing

Spatial Domain Processing

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Spatial domain: Image plane itself, directly process the intensity

values of the image plane.

Transform domain (Frequency domain, Wavelet domain) Process

the transform coefficients, not directly process the intensity values

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Gray level transform

Intensity Level slicing

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Spatial domain refers to the aggregated of pixels composing an

: an operator defined over some neighborhood of ( ,  ), maybe

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Normalized histogram: An estimate of the probability of

occurrence of gray level

The sum of all components of a normalized histogram is equal to 1.

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

Dr. Che Viet Nhat Anh

Digital Image & Speech Processing

could have the same histogram.

Digital Image & Speech Processing

image processing applications:

Histograms are simple to calculate in software & also lend

: an operator defined over some neighborhood of ( , ), maybe

depends only on the value of at , . becomes a gray-level