Anda di halaman 1dari 22

1.

INTRODUCTION
1.1 Infrared Thermography
Infrared (IR) thermography is a branch of imaging science that detects
electromagnetic radiation emitted by objects in infrared range and generates the
corresponding radiation images, called thermograms. As all objects with temperature above
absolute zero, or -273.15 degree Celsius (C), emit infrared radiation from their surface
(Bronzino, 2005). There are actually two types of infrared imaging system, namely active
infrared and passive infrared. Active infrared requires targets to be illuminated by near
infrared beams and the reflected radiation is measured. However, active infrared system is
greatly limited by the distance between infrared source and the target. Meanwhile, passive
system is sensitive towards far infrared wavelengths, or more commonly known as heat
radiation, dissipated by the target itself. Since the thermal imaging systems do not depend
on reflected ambient light, the generated thermal images are entirely ambient light-level
independent.
The amount of radiation emitted by an object increases as its temperature rises.
Using a thermographic imaging system, the variations in infrared radiation emitted by an
object of interest along with its surrounding can be approximated into analogous
temperature values and produces the corresponding thermogram. Stefan-Boltzmann Law
states that the total emissive power of an object is proportional to the emissivity of its
surface and the fourth power of its absolute temperature (Bronzino, 2005).
4 2
[ / ]
i i
E T Watt m
(1.1)
where E
i
is the total emissive power;

i
is the emissivity;

is the Stefan Boltzman constant;


T is the absolute temperature of an object.
Wavelengths outside the visible electromagnetic spectrum do not map uniformly
into the colour system adopted by human vision, therefore the significance of colours
become lesser. Thus, infrared cameras often generate monochromatic thermograms as a
mean to reduce the hardware complexity. In other words, different temperatures of objects
are distinguished with different shades of grey level. A high greyscale level indicates a
high temperature object which appears to be a bright object on the infrared image and vice
versa. Contrast seen on the image represents the difference of temperature between the
objects. However, as human eyes only capable of distinguish few dozen shades of grey
(Gonzalez & Woods, 2002), the risk of neglecting vital information contents while
analysing monochromatic images is relatively high. Moreover, many infrared imaging
systems often produce infrared images with low signal-to-noise ratio (SNR) and poor
contrast (Ni, Li, & Xia, 2008), which further impedes the proper investigation of
information contents. Therefore, contrast enhancement becomes an essential key step in
infrared imaging based applications.
1.2Breast Infrared Thermography
Examination of heat pattern and vessels development in breast is the very first
application of infrared imaging in medicine (Qi & Diakides, 2007). This imaging technique
is known as breast thermography.
The underlying biological principle lies in the development of new blood vessel in
the breast during precancerous state in a process called angiogenesis. Malignant
angiogenesis is a process where cancerous cells induce the development of new blood
vessels in order to enhance the mass transfer of nutrient and waste to the cancerous cells
(Wang et al., 2010). It is found that malignant angiogenesis starts long before the cells turn
cancerous (Bronzino, 2005). There is a rise in infrared radiation due to higher metabolic
activity and blood flow at the lesion site (Wang et al., 2010). Therefore, detection of this
malignant angiogenesis is the key tool for early detection of breast cancer. Malignant
angiogenesis is characterised by a raise in vascular asymmetry and heat observable through
infrared images. In other words, early stage of breast cancer is detectable through
examination of heat pattern change of breast thermogram.

Figure 1.1: Colourised Infrared Image of Breast (Hobbins & Amalu, 2011)
Figure 1.1 shows a comparison of colourised normal breast image with abnormal
breast image. The picture on the left is an infrared image of normal breast. It is
characterised by dark colour that indicates the low-heat/balanced cool area in breast region.
Meanwhile, the right picture shows an image of abnormal breast, typified by high heat-
activity in the left breast due to formation of new blood vessel for cancerous cells survival.
In short, breast thermography is capable of giving an early sign of possible
development of breast cancer up to 10 years in advance of any other imaging tool. Infrared
imaging technique is emerging to first-line screening tool to both clinical diagnosis and
health-maintaining thanks to significant advancement in infrared sensing technology as
well as improvement in image-processing algorithm (Qi & Diakides, 2007). Furthermore, a
high detection rate of 95% in early stage cancer is achieved with integration of breast
thermography into mammography and clinical examination (Bronzino, 2005).
2. RESEARCH PROBLEM
Most infrared cameras generate monochromatic thermograms that reflect
temperature differences with different shades of grey intensity. Consider an image I(x,y)
that consists of L discrete grey intensity levels; for a 8-bit greyscale image, L equals to 256,
which means there are a total of 256 different shades to represent the temperatures.
Contrast seen on the image represents the difference of temperature between the objects.
However, as human eyes only capable of distinguish few dozen shades of grey (Gonzalez
& Woods, 2002), the risk of neglecting vital information contents while analysing
monochromatic images is relatively high. Moreover, many infrared imaging systems often
produce infrared images with low signal-to-noise ratio (SNR) and poor contrast (Ni, Li, &
Xia, 2008), which further impedes the proper investigation of information contents.
Therefore, contrast enhancement becomes an essential key step in infrared imaging based
applications.
2.1 Problem Statement
Infrared thermography screening provides a viable solution for early cancer
detection by identifying abnormal heat dissipated due to formation of malignant
angiogenesis. However, quality of infrared image is greatly degraded by low contrast and
low signal to noise ratio. Furthermore, the relative small heat difference between mutated
cells and normal cells, especially during early formation, may be difficult to differentiate
by inspecting the low contrast raw infrared image (Kennedy & Seely, 2009). Hence, it is
crucial to apply image contrast enhancement to improve the image vision quality of the
infrared image. This allows clinician to easily identify the presence of breast lesion and the
enhancement process can serves as a pre-emptive step to make the infrared image more
suitable for further analysis procedures, such as pattern recognition or to automatically
identify the lesions without human intervention.
3. OBJECTIVES
To develop an improved contrast enhancement algorithm to overcome or mitigate
the limitations of existing methods.
To develop a stand-alone image enhancement algorithm for breast infrared image.
4. HYPOTHESIS
Better contrast enhancement technique will help to magnify the subtle temperature
differences and make them more noticeable; thus ease the effort and improve accuracy of
clinician in identifying early formation of malignant angiogenesis in breast region.
5. LITERATURE REVIEW
5.1 Image Contrast Enhancement Techniques
Medical images are images that require high-contrast for a right diagnosis. The
most essential aspect of medical image analysis lies in the precise definition of subjects
contours which requires high contrast between objects. (Larrabide, Novotny, Feijo, &
Taroco, 2005). However, infrared imaging tools often generate low-contrast images and
this downside reduces the detectability of small temperature difference present in the image
(Zhan & Wu, 2010). There are two main categories of enhancement techniques namely
spatial domain technique and frequency domain technique. Spatial domain technique deals
with the image pixels while frequency domain technique operates on the Fourier-
transformed of the image (Ahmed & Nordin, 2011). For spatial method, simplicity of
Histogram equalisation contributes to its extensive employment in contrast enhancement.
However, Histogram equalisation suffers from the problem of over-enhancement and
image deterioration. A number of techniques have been used to overcome its annoying
effects. Nevertheless, each technique has its advantages and drawbacks (Bansal & Goyal,
2011).
5.2.1 Histogram Equalisation (HE) Technique
Histogram Equalisation (HE) is one of the fundamental and widely used image
processing techniques for improving the contrast of an image via dynamic range
modification (Gonzalez & Woods, 2002). In HE, the histogram of an image is redistributed
and forced to be uniform by flattening its distribution of intensities according to a desired
transfer function (Andrew, Tescher, & Kruger, 1972).
Consider a two dimensional image I(x,y) that consists of L discrete gray intensity
levels; for a 8-bit greyscale image, L equals to 256. The probability density function (PDF),
pdf(k), can be defined as (Yoon, Han, & Hahn, 2009)
( ) ; 0
k
n
pdf k k L
n
< ,
(5.1)
where n
k
is the number of pixels for k
th
grey level, while n denotes the total number of
pixels for input image. Using the PDF, the corresponding cumulative distribution function
(CDF), cdf(k), can be defined as
( )
0
( ); 0
k
i
i
cdf k pdf k k L

<
. (5.2)
Based on the CDF, equalisation can be carried out by modifying the CDF of input image
using a uniform transfer function as following (Pratt, 2007)
( ) ( )
max min min
, , ( ) ; 0 I x y k k k cdf k k k L + <
,
(5.3)
where I(x,y,k) denotes the enhanced output image with modified probability distribution for
intensity value of k.
Figure 5.1 (a) and (c) show an example of infrared image and the result after HE,
while Figure 5.1 (b) and (d) illustrate their corresponding histograms, respectively. It can
be clearly observed that the dynamic range of the output image is being stretched,
especially around the peak regions to achieve more uniform distribution.
(a) (b)
(c) (d)
Figure 5.1: Histogram Equalization (HE) Technique. (a) Original Breast Thermogram, (b)
Histogram of Original Breast Thermogram, (c) Enhanced Image after HE, and (d)
Histogram of Enhanced Image after HE.
5.2.2 Advantages and Disadvantages
As shown in Figure 5.1, HE greatly improves the image vision quality of the
original low contrast infrared image by expanding its dynamic range over the entire
available intensity levels; image details previously obscured by the low contrast value are
made perceptually visible. However, to achieve an uniform probability density, a much
higher contrast gain (change of grey level) will be given to peak regions of the histogram
compared to the rest; in short, contrast gain is scaled proportional to the height of the
histogram. Such phenomenon often leads to drastic brightness change of the input image,
causing artefact known as over-equalisation (Chang & Wu, 1998; Bansal & Goyal, 2011).
5.3.1 Brightness Preserving Bi-Histogram Equalization (BBHE) Technique
Brightness Preserving Bi-Histogram Equalization (BBHE) is a variation of HE with
the aim to minimize the over-equalization effect via preserving the original image mean
brightness. This can be achieved by performing mean-separation algorithm (Kim, 1997);
histogram of image is divided into sub-histograms based on the original image mean value
and equalised independently. The separating point, X
m
, can be calculated as
( ) { }
1
0
; 0
L
m
k
X k pdf k k L

<
, (5.4)
where pdf(k) is the PDF at k
th
grey intensity level; L is the total number of available grey
intensity levels. Using X
m
to separate the histogram, the PDF and CDF for lower and upper
sub-histograms can be written as
( ) ;
k
l
l
n
pdf k
n


( )
0
( ) ; 0
k
l l i m
i
cdf k pdf k k X

,
( ) ;
k
u
u
n
pdf k
n


( )
1
( ); 1
m
k
u u i m
i L
cdf k pdf k X k L
+
+ <

,
(5.5)
(5.6)
where pdf
l
and pdf
u
are the PDF, n
l
and n
u
are the total pixels, cdf
l
and cdf
u
are the CDF for
the lower and upper sub-histograms, respectively.
By applying the uniform transfer function shown in Equation (5.3), the both sub-
histograms are equalised independently and recombined to produce the complete transfer
function for entire image. In other words, lower region is equalised within the range of zero
to the mean, while the upper region is equalised within the range of mean to last level
(Bansal & Goyal, 2011). Figure 5.2 (a) shows an example of infrared image enhanced by
BBHE.
5.3.2 Advantages and Disadvantages
By dividing the image histogram into sub-histograms, BBHE allows full utilisation
of available dynamic range with lesser over-enhancement artefact. However, the mean
preservation principle of BBHE is based on the assumption that the histogram has
symmetrical distribution around the separating point, X
m
; the probability function, p, of
both sub-histograms is equal. The mean brightness of the enhanced image, X
out
, can be
written as
( )
0 1
1
2 2
m m L
out
X X X X
X p p

+ + _ _
+

, ,
, (5.7)
where X
0
is the minimum intensity level; X
L-1
is the maximum intensity level. Yet in reality,
especially for medical image with large section of uniform background, the histogram
distribution is often skewed towards either side depending on the background colour; the
mean brightness value of enhanced image tends to incline towards region with higher pixel
density, hence results in a significant shift from the original mean brightness value (Wang,
Chen, & Zhang, 1999). Such phenomenon reduces the capability of BBHE to restrain the
over-equalisation artefact.
5.4.1 Dualistic Sub-Image Histogram Equalization (DSIHE) Technique
Dualistic Sub-Image Histogram Equalization (DSIHE) is similar to BBHE except
that DSIHE separates the histogram based on its CDF rather than the mean brightness value
(Wang et al., 1999); the histogram is divided at middle point of CDF (CDF = 0.5), thus the
total pixels (or area) for both sub-histograms is always equal and symmetrically distributed
around the separating point. The probability function, p, of each sub-histogram is always
equal to 0.5; in other words, DSIHE is an ideal case for BBHE.
By substituting p equals to 0.5 into Equation (5.7), the mean brightness value for enhanced
image, X
out
can be derived as
2
m g
out
X X
X
+
, (5.8)
(5.9)
( )
0 1
0.5
g L
X X X

+
,
where X
g
is the middle grey level intensity. Figure 5.2 (b) shows an example of infrared
image enhanced by DSIHE.
5.4.2 Advantages and Disadvantages
DSIHE successfully eliminates the drawback of BBHE as the histogram will
always be symmetrically distributed around the separating point. However from Equation
(5.8), it can be observed that both X
m
and X
g
have same influence on the value of X
out
. Such
level of brightness preservation may be insufficient to restrain the over-enhancement
artefact caused by the presence of large section of uniform background in medical images
(Chen & Ramli, 2003); the greater the skewness of histogram distribution, the larger the
deviation between value of X
m
and X
g
, the poorer the capability of DSIHE to restrain the
over-enhancement artefact due to greater shift of output mean brightness from the original
mean brightness value.
5.5.1 Recursive Mean-Separate Histogram Equalization (RMSHE) and Recursive
Sub-Image Histogram Equalization (RSIHE) Techniques
As the performance of BBHE and DSIHE are greatly limited by the skewness of
histogram distribution, recursive approach for both techniques were suggested; Chen and
Ramli (2003) proposed Recursive Mean-Separate Histogram Equalization (RMSHE) and
Sim, Tso, and Tan (2007) proposed Recursive Sub-Image Histogram Equalization (RSIHE)
for BBHE and DSIHE, respectively. The separating point of recursive approaches, namely
X
RMSHE(j)
and X
RSIHE(j)
, can be defined as
( )
( ) { }
1
0
; ; 1 2
RMSHE
L
r
j min max j
k
X k pdf k l k l j


,
( )
( ) , 0.5;
RSIHE
r
j j
X k if cdf k 1 j 2
,
(5.10)
(5.11)
l
min
and l
max
are the minimum and maximum intensity levels for corresponding j
th
sub-
histogram; r is a positive integer value that represents the recursive level.
To simplify the notation, let all the sub-histograms be symmetrically distributed
around their separating point so that both X
RMSHE(j)
and X
RSIHE(j)
will always equal, the mean
brightness value for enhanced image, X
out
, of each recursive level can be written as
(2 1)
;
2
r
m g
out
r
X X
X 0 r
+
<
,
;
2
g m
out m
r
X X
X X 0 r

+ < .
(5.12)
From Equation (5.12), as the value of r increases, the value of X
out
will eventually converge
to original mean brightness value, X
m
. Figure 5.2 (c) and (d) show examples of infrared
image enhanced by RMSHE and RSIHE, respectively.
5.5.2 Advantages and disadvantages
Recursive approaches provide better brightness preservation thus capable to exert
higher degree of restriction of over-enhancement effect. However, as the level of recursive
increase, the degree of contrast enhancement will decrease as smaller dynamic range for
each sub-histogram will restrict the contrast gain. Furthermore, there is no governing rule
to indicate suitable recursive level for optimized enhancement effect (Yoon et al., 2009).
(a) (b)
(c) (d)
Figure 5.2: Sub-histograms Enhancement. (a) Image Enhanced by BBHE, (b) Image
Enhanced by DSIHE, (c) Image Enhanced by RMSHE, and (d) Image Enhanced by
RSIHE.
5.6.1 Plateau Histogram Equalisation (PHE) Technique
Plateau Histogram Equalization (PHE) is another variation of HE which attempts to
reduce the over-enhancement artefact by restraining the height of peak regions of
histogram (Vickers, 1996). As mentioned in previous discussion, the contrast gain is highly
associated with height (or density) of histogram; hence by lowering the height of high peak
regions, the aforementioned over-enhancement artefact can be reduced.
To restrict the height of peak regions, Vickers (1996) suggested applying a plateau
threshold value, T, globally over the entire histogram; any of the histogram counts that
have value larger than T will be truncated. The thresholding procedure can be
represented by following equation
( )
( ) ( )
( )
( )
,
; 0 ; 1
,
T max
pdf k pdf k T
pdf k k L T pdf k
T pdf k T
<

< < <


'

,
(5.13)
where pdf
T
(k) and pdf(k) are the modified (or truncated) PDF and original PDF for k
th
intensity level, respectively; pdf
max
(k) is the maximum PDF value. Finally, the enhanced
output image can be obtained by solving Equation (5.13) with both Equation (5.2) and
(5.3). Figure 5.3 (a) shows an example of infrared image enhanced by PHE.
5.6.2 Advantages and Disadvantages
With properly selected plateau threshold value, PHE able to restrain the over-
enhancement effect while achieve adequate contrast enhancement. However, the plateau
threshold value for PHE is an arbitrarily value that limits its practical usefulness as the
nature of image may vary from one to another (Wang, Liu, Li, & Zhou, 2006); there is no
governing rule to suggest a suitable threshold value which can adapt to different types of
image.
5.7.1 Self-Adaptive Plateau Histogram Equalisation (SAPHE)
To resolve the drawback of PHE, Wang et al. (2006) proposed a variation of PHE,
known as Self-Adaptive Plateau Histogram Equalisation (SAPHE) that capable of deriving
a self-adapted plateau threshold value for each image. For SAPHE, the median value of
local maxima is selected as plateau threshold value; the local maxima will always stay in
between the value of one and the highest histogram counts, hence it fully complies with the
principle of PHE.
Let h(k) be the histogram function and the slope of histogram function, h(k), can be
derived as
( ) ( ) ( ) ' 1 ; 0 1 h k h k h k k L < <
,
(5.14)
A local maximum happens when the slope reaches positive and later fall to negative as
( ) ( ) ' 1 0, , ' 1 0 h k Max h k > + <
.
(5.15)
Once all the local maxima are identified, the local maximum with largest value (or global
maximum value) is excluded and the plateau threshold value is estimated by finding the
median value of the remanding local maxima. Once the plateau threshold value is found,
the image histogram is truncated using Equation (5.13), followed by CDF modification and
equalisation to generate the desired enhanced output image. Figure 5.3 (b) shows an
example of infrared image enhanced by SAPHE.
5.7.2 Advantages and Disadvantages
For SAPHE, the plateau threshold value can adapted to different types of image as
the threshold is derived from the image histogram itself and no human intervention is
needed. However, there is no any indication of relationship between better contrast
enhancement and local maxima value, thus the performance of SAPHE is rather
inconsistent.
(a) (b)
Figure 5.3: Plateau Histogram Enhancement. (a) Image Enhanced by PHE (threshold =
200), (b) Image Enhanced by SAPHE.
6. METHODOLOGY
In previous discussion, all the discussed methods are considered as global
enhancement; the overall contrast of an image is improved without taken the local image
details into consideration. For better vision quality, local enhancement is preferable as it is
difficult to simultaneously enhance the contrast of different regions with a single
transformation (Zhu, Chan, & Lam, 1999). Therefore in this study, a local enhancement
method known as Adaptive Double Plateau Equalisation (ADPE) is suggested.
For ADPE, the principle of PHE is used due to its simplicity and satisfactory
enhancement result. Unlike conventional PHE, ADPE utilises two plateau threshold values;
the upper plateau threshold for minimising the over-enhancement artefact, while lower
plateau threshold for improving the contrast gain for subtle image details. The ADPE
consists of three stages; image localisation, local histogram modification and image
reconstruction.
First stage, the image localisation; the input image is divided into arbitrarily
number of equally-sized sub-regions, as shown in Figure 6.1.
(a)
Figure 6.1: Sub-regions after Image Localisation.
Second stage, the local histogram modification; the upper and lower plateau
thresholds are used to modify the local histogram. As mentioned earlier, over-enhancement
artefact is caused by the inequality of histogram distribution and a general measure for such
inequality is Lorenz curve. By describing the proportional distribution between the
cumulative histogram counts and the number of available intensity levels with Lorenz
curve, optimal plateau thresholds can be estimated. To plot the curve, all the input
parameters must first be ordered in ascending manner. Figure 6.2 shows an example of
Lorenz curve.
Once the curve is plotted, the inequality of a local histogram, , can be derived as
( ) ( ) { }
( )
1
0
1
0
-
L
e l
k
L
e
k
U k U k
area of concavity
area under line of equality
U k

, (6.1)
where U
e
(k) and U
l
(k) are the magnitude of line of equality and Lorenz curve for k
th
available intensity level, respectively. Using Equation (6.1), the upper and lower plateau
threshold, T
u
and T
l
, can be derived as
( ) 1
u max l
T H T +
,
l
n
T
L
,
(6.2)
(6.3)
where H
max
is the highest local histogram counts; n is the total number of pixels within the
local sub-region; L is the total number of intensity levels.
During the histogram modification, the histogram counts are truncated by the upper
plateau threshold. The extra counts, E, that being clipped off are uniformly redistributed
among intensity levels with counts lesser than the lower plateau threshold; the purpose of
redistribution is to improve the contrast gain for subtle image details. The redistribution
of extra counts can be written as
( ) ( ) , ( ) ;
l
E
h k h k if h k T 0 k L
N
+ < < ,
(6.4)
where N is the total number of low-counts intensity level.
(a) (b)
Figure 6.2: Lorenz Curve. (a) Image Histogram. (b) Lorenz Curve.
Third stage, the image reconstruction; image is reconstructed by equalising each
sub-region independently with corresponding modified local histogram distribution.
However as the equalisation is performed locally, undesired boundary effect due abrupt
change of intensity values may be noticed near the region boundaries. To minimise such
artefact, bilinear-interpolation is performed onto the enhanced image; new intermediate
intensity values are estimated using sets of known intensity values extracted from adjacent
sub-regions based on the ratio of their spatial distance. By integrating the newly estimated
intermediate intensity values into the enhanced image, the block effect can be smoothed-
out and the desired output image with minimal artefact can be obtained.
7.0 EXPECTED RESULT
An important step in assessing the quality of enhancement for contrast
enhancement methods is to find a suitable quantitative performance measure. One of the
commonly used performance measure for contrast enhancement is known as Michelson
Contrast Measure (AME) (Agaian, Silver, & Panetta, 2007). For AME, the performance of
contrast enhancement is quantified by measuring the relationship between the difference
and the sum of lowest and highest intensity values that found within a sub-image, followed
by taking the average value of each block over the entire image. Let the input image be
decomposed into a total of (k
1
k
2
) blocks of B
m,n
, the AME can be written as
1 2
1 1
1 2
1
ln
k k
max;m,n min;m,n max;m,n min;m,n
m n
max;m,n min;m,n max;m,n min;m,n
I I I I
AME
k k I I c I I c

_ _



+ + + +
, ,
,
(7.1)
where I
max;m,n
and I
min;m,n
are the maximum and minimum intensity values within the block
B
m,n
, respectively; c is a small constant equals to 0.001 to avoid dividing by zero. It is
expected that ADPE is capable of producing enhanced infrared image with AME measure
higher than rest of the conventional methods.
8.0 BUDGET
Quantity Description Unit Price (RM) Total (RM)
1 Fixed Infrared Camera:FLIR SC305 29,600.00 29,600.00
1 MathWorks MatLab R2010b 6,200.00 6,200.00
Total 35,800.00
9.0 CONCLUSION
In this study, several conventional image contrast enhancement methods have been
discussed in details, along with their advantages and disadvantages. A novel local contrast
enhancement method, known as Adaptive Double Plateau Equalisation (ADPE) is
proposed to overcome or mitigate the aforementioned drawbacks of the conventional
method. Furthermore, a quantitative measure of quality of contrast enhancement is
suggested to quantify the performance of all discussed methods. It is strongly believed that
the performance of ADPE will surpass all the conventional methods.
10. REFERENCES
[1] Agaian, S. S., Silver, B., & Panetta, K. A. (2007). Transform coefficient histogram-
based image enhancement algorithms using contrast entropy. Institute of Electrical
and Electronics Engineers Transactions on Image Processing, 16(3), 714-758.
[2] Ahmed, H.S.S., & Nordin, M.J. (2011). Improving diagnostic viewing of medical
images using enhancement algorithm. Journal of Computer Science, 7(12), 1831-
1838.
[3] Andrews, H. C., Tescher, A. G., & Kruger, R. P. (1972). Image processing by digital
computer. Institute of Electrical and Electronics Engineers Spectrum, 9(7), 20-32.
[4] Bansal, V., & Goyal, B. (2011). Natural gray scale image contrast enhancement using
histogram equalization technique. International Journal of Computing and Corporate
Research India, 1(1), 143-157.
[5] Bronzino, J. D. (2005). The biomedical engineering handbook: Medical devices and
systems, 3
rd
eds. New York: CRC Press.
[6] Chang, D. C., & Wu, W. R. (1998). Image contrast enhancement based on a
histogram transformation of local standard deviation. Institute of Electrical and
Electronics Engineers Transactions on Medical Imaging, 17(4), 518-531.
[7] Chen, S.D., & Ramli, R. (2003). Contrast enhancement using recursive mean-
separate histogram equalization for scalable brightness preservation. Institute of
Electrical and Electronics Engineers Transactions on Consumer Electronics, 49(4),
1301-1309.
[8] Gonzalez, R. C., & Woods, R. E. (2002). Digital Image Processing, 2
nd
eds. New
Jersey: Prentice Hall.
[9] Hobbins, W., & Amalu, W. (2011). Beating breast cancer. Retrieved 25 October
2011, from http://www.medithermclinic.com/breast/breast.html
[10] Kennedy, D., & Seely, D. (2009). A comparative review of thermography as a breast
screen technique. Integrated Cancer Therapies, 8(01), 9-16.
[11] Kim, Y. T. (1997). Contrast enhancement using brightness preserving bi-histogram
equalization. Institute of Electrical and Electronics Engineers Transactions on
Consumer Electronics, 43(1), 1-8.
[12] Larrabide, I., Novotny, A. A., Feijo, R. A., & Taroco, E. (2005). A medical
enhancement algorithm based on topological derivative and anisotropic diffusion.
Paper presented at the XXVI Iberian Latin-American Congress on Computational
Methods in Engineering, Espirito Santo, Brazil.
[13] Ni, C., Li, Q., & Xia, L. Z. (2008). A novel method of infrared image de-noising and
edge enhancement. Signal Processing, 88, 1606-1614.
[14] Pratt, W. K. (2007). Digital image processing 4
th
eds. California: Wiley Interscience.
[15] Qi, H., & Diakides, N. A. (2007). Infrared imaging in Medicine. Baco Raton: CRC
Press.
[16] Sim, K. S., Tso, C. P., & Tan, Y. Y. (2007). Recursive sub-image histogram
equalization applied to gray scale images. Pattern Recognition Letter, 28, 1209-1221.
[17] Vickers, V. E. (1996). Plateau equalization algorithm for real-time display of high-
quality infrared imagery. Society of Photo-Optical Instrumentation Engineers, 35(7),
1921-1926.
[18] Wang, B. J., Liu, S. Q., Li, Q., & Zhou, H. X. (2006). A real-time contrast
enhancement algorithm for infrared images based on plateau histogram. Infrared
Physics & Technology, 48, 77-82.
[19] Wang, J., Chang, K. J., Chen, C. Y., Chien, K. L., Tsai, Y. S., Wu, Y. M., et al.
(2010). Evaluation of the diagnostic problem of infrared imaging of the breast: A
preliminary study. Bio-Medical Engineering Online, 9(3), 1-10.
[20] Wang, Y., Chen, Q., & Zhang, B. M. (1999). Image enhancement based on equal area
dualistic sub-image histogram equalization method. Institute of Electrical and
Electronics Engineers Transactions on Consumer Electronics, 45(1), 68-75.
[21] Yoon, H. S., Han, Y. J., & Hahn, H. S. (2009). Image contrast enhancement based
sub-histogram equalization technique without over-equalization noise. World
Academy of Science, Engineering and Technology, 50, 176-182.
[22] Zhan, B., & Wu, Y. (2010). Infrared image enhancement based on wavelet
transformation and retinex. IHMSC 2010: 2010 Second International Conference on
Intelligent Human-Machine Systems and Cybernetics: Proceedings of the 2nd
International Conference on Intelligent Human-Machines Systems and Cybernatics
held on 26 - 28 August 2010 at the Central Hotel, Nanjing (pp. 313-316). Nanjing: In
stitute of Electrical and Electronics Engineers.
[23] Zhu, H., Chan, F. H. Y., & Lam, F. K. (1999). Image contrast enhancement by
constrained local histogram equalization. Computer Vision and Image
Understanding, 73(2), 281-290.

Anda mungkin juga menyukai