Beruflich Dokumente
Kultur Dokumente
IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 15, NO. 10, OCTOBER 2006
I. INTRODUCTION
MAGE enhancement techniques are used to improve the appearance of the image or to extract the finer details in the
degraded images. Color image enhancement using RGB color
space is found to be inappropriate as it destroys the color composition in the original image.
Image enhancement can be treated as transforming one image
to another so that the look and feel of an image can be improved
for machine analysis or visual perception of human beings. For
grayscale image enhancement, the most popular method is histogram equalization, which is based on the assumption that a
uniformly distributed grayscale histogram will have the best visual contrast. Some other methods are the variants of histogram
equalization. However, generalizing grayscale image enhancement to color image enhancement is not a trivial task. Several
factors, such as selection of a color model, characteristics of
the human visual system, and color contrast sensitivity, must be
considered for color image enhancement.
Manuscript received May 30, 2004; revised December 15, 2006. The associate editor coordinating the review of this manuscript and approving it for publication was Dr. Eli Saber.
M. Hanmandlu is with the Indian Institute of Technology, Delhi, New Delhi
110016, India (e-mail: mhmandlu@ee.iitd.ac.in).
D. Jha is with Scientific Analysis Group, Defence R&D Organisation,
Metcalfe House, Delhi 110054, India (e-mail: devendrajha@indiatimes.com).
Color versions of Figs. 316 are available online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TIP.2006.877499
HANMANDLU AND JHA: OPTIMAL FUZZY SYSTEM FOR COLOR IMAGE ENHANCEMENT
2957
(1)
where
or
represents the membership or
of
,
is the color intensity
grade of some property
at
pixel. For a color image, the membership functions
. For the
are taken for the unions of all colors
transformation of the color in the range (0255) to the fuzzy
property plane in the interval (0,1), a membership function of
the Gaussian type
(2)
is suggested in [13] and it contains a single fuzzifier, . Here,
is the maximum color value present in the image,
being the number of levels of intensity. Though valid, for
under exposed images, this needs to be modified for under plus
over exposed images. Thus, this is also a design parameter as
will be discussed in the Section VI.
, with
The membership values are restricted to the range
. For computational efficiency, histogram of
is considered for fuzzification. So,
represents
color
for a value , with
the membership function of color
, defined by
(3)
replaced by the
This function is the same as in (2), with
index , the intensity of the color components having the fre. As outlined in the introduction, we
quency occurrence,
to modify its membership function without
will consider
disturbing the membership functions of other two components.
are higher for a brighter image.
It is observed that values of
The membership values are transformed back to the spatial domain after the desired operator is applied in the fuzzy domain.
2958
IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 15, NO. 10, OCTOBER 2006
by
(8)
(4)
and are the modified membership function and
where
intensity value, respectively.
We restrict the enhancement of the image by whatever is possible with the fuzzy contrast intensification operator. The original contrast intensification operator, INT [4] depends on the
membership function only. It needs to be applied successively
on an image for obtaining the desired enhancement. This limitation is removed in the new intensification (NINT) operator
proposed in [13]. This is a parametric sigmoid function given
by
(5)
Here, we propose a general intensification operator (GINT) by
replacing 0.5 by
in (5) leading to
(9)
In the above definition, the fuzzy average contrast gives the
overall intensity of the image whereas the fuzzy contrast gives
the spread of the gradient with respect to the reference (the
cross over point). Their ratio is found to give the quality of the
image. The amount of enhancement will be indicated by the visual factor to be defined later.
by
in (7) and (8), we obtain the
If we replace
fuzzy contrast and average contrast for the original image. These
are given by
(10)
(6)
The unknown parameters in the above equation are intensification parameter, and crossover membership function or
simply crossover point, .
(11)
In view of the above definition, the image quality of the original
image is given by
(12)
The change in image quality is now given by
(13)
(14)
(7)
where
HANMANDLU AND JHA: OPTIMAL FUZZY SYSTEM FOR COLOR IMAGE ENHANCEMENT
the parameters ,
and
spect to ,
with re-
2959
In view of (23), we can combine (20) and (22) into the following:
(24)
(15)
Then (24) can be written as
(16)
(25)
The change in the image quality is given by
(17)
(26)
where
is defined in Appendix I as (A15). The above
and
derivatives are used in the learning of the parameters ,
. Before presenting the algorithm for learning, we need to
introduce certain constraints for generating aesthetic images.
These constraints are discussed in Section IV. The initial value
is taken from [15]
of
(18)
(27)
(28)
(29)
The optimization of the above objective functions is given in
Appendix I.
as
IV. VISUAL FACTORS
(19)
by
(21)
would simplify (16) as
(22)
Now, the entropy-based quality factor,
easily follows from
.
the definition of
Definition: The entropy-based quality factor of an image is
defined as the ratio of entropy-based average fuzzy contrast
to the entropy-based fuzzy contrast,
(23)
The problem at hand is optimization of the entropy with respect to the parameters , and with certain constraints. For
a solution of this problem, we adapt the univariate method of
[18] by multiplying the step size, by the derivative,
of the objective function with respect to the base point, .
In this method, we change only one parameter at a time, and
produce a sequence of improved approximations to reach the
minimum point. By starting at a base point
in the th iteration, we fix the values of any one of
parameters and vary the remaining parameter. The purpose is to
. The search is now continued in
produce a new base point
2960
IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 15, NO. 10, OCTOBER 2006
and set
as
. If
,
will be the correct
,
direction for decreasing the value of , and if
will be the correct direction. If both
and
are less than
, we take
as the minimum of the two.
;
4) Set
5)
6) Set
and go to Step 2). Continue this procedure
until no significant change is observed in the value of the
objective function.
We have taken a unit step length for computational simplicity.
The flow chart detailing the enhancement technique is shown in
Fig. 1 and the algorithm for the same is as follows.
Algorithm for Image Enhancement
..
.
where
using (18).
4) Fuzzify
5) Initialize
to get
using (3).
, and calculate
, and
HANMANDLU AND JHA: OPTIMAL FUZZY SYSTEM FOR COLOR IMAGE ENHANCEMENT
TABLE I
INITIAL PARAMETERS OF TEST IMAGES
TABLE II
OPTIMIZATION OF E WITH Q
= 0:4, = 0:2
6) Initialize
2961
TABLE III
OPTIMIZATION OF E WITH Q
= 0 :4 , = 0 :2
TABLE IV
OPTIMIZATION OF E WITH C = 0:4, = 0:3
2962
Fig. 2. Top to bottom: (a) Original image and (b) enhanced image of doctor.
Fig. 3. Top to bottom: (a) Original image and (b) enhanced image of face.
IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 15, NO. 10, OCTOBER 2006
Fig. 4. Top to bottom: (a) Original image and (b) enhanced image of timber.
depends on the relative dominance of one over another. For example in Fig. 15 (the cougar) the under exposed portion dominates the over exposed portion. Hence, it is treated as the under
exposed, thus giving the visual factor below 1.
The proposed approach works better; in other words it is well
suited for the enhancement of under exposed images because of
the choice of the fuzzification function, in which we are taking
the difference of gray levels with respect to maximum gray level
and because of the property of global intensification operator.
HANMANDLU AND JHA: OPTIMAL FUZZY SYSTEM FOR COLOR IMAGE ENHANCEMENT
2963
Fig. 5. Top to bottom: (a) Original image and (b) enhanced image of lab.
Fig. 6. Top to bottom: (a) Original image and (b) enhanced image plane.
Fig. 9. Top to bottom: (a) Original image and (b) enhanced image Lena.
Fig. 7. Top to bottom: (a) Original image and (b) enhanced image of meeting.
Fig. 10. Top to bottom: (a) Original image and (b) enhanced image of water.
Fig. 8. Top to bottom: (a) Original image and (b) enhanced image of girl.
for the optimization. We have also introduced quality and visual factors as constraints in the optimization of entropy. A visually pleasing image is obtained with the appropriate choice of
quality factors. It may be noted that GINT is guided by these
factors since ultimate enhancement leads to the binarization of
the image. The results of enhancement using fuzzy entropy optimization are compared with those of histogram equalization.
Fig. 11. Comparison of enhancement of fruit image: (a) Original, (b) proposed
approach, and (c) histogram equalization.
For over-exposed and mixed-type images, the maximum intensity of the color in the Gaussian fuzzification function becomes another parameter in addition to the three parameters.
2964
IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 15, NO. 10, OCTOBER 2006
Fig. 12. RGB histograms for fruit image: (a) Original, (b) proposed approach, and (c) histogram equalization.
Fig. 13. RGB histograms for fruit image: (a) Original, (b) proposed approach, and (c) histogram equalization.
Fig. 14. RGB histograms for fruit image: (a) Original, (b) proposed approach, and (c) histogram equalization.
+ over exposed cougar image: (a) Original, (b) proposed approach, and (c) histogram equalization.
Fig. 16. Comparison of enhancement of over exposed natural scene: (a) Original, (b) proposed approach, and (c) histogram equalization.
HANMANDLU AND JHA: OPTIMAL FUZZY SYSTEM FOR COLOR IMAGE ENHANCEMENT
2965
TABLE V
HISTOGRAM EQUALIZATION RESULTS WITH = 0:5
(A9)
(A10)
(A11)
The derivatives in (A9)(A11) are as follows:
(A12)
We have not determined this by optimization as this parameter has been manipulated by visual assessment without going
through the rigors of optimization procedure. The maximum
value is decreased for mixed type image and increased for over
exposed images. The visual factor provides a clue for knowing
the type of the image. However, this information is available
after enhancement.
APPENDIX I
(A13)
(A14)
but
(A15)
(A16)
Case I: Optimization of With the Image Quality as the Constraint: For this, consider the objective function
(A17)
(A1)
where
(A18)
(A4)
(A20)
(A5)
is the desired image quality. Differentiating
where
with respect to the unknown parameters yields the following
derivatives:
(A6)
(A21)
where
(A7)
(A22)
(A8)
(A23)
2966
IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 15, NO. 10, OCTOBER 2006
(A24)
(A25)
Case III: Optimization Using the Fuzzy Contrast: This definition of fuzzy contrast gives one way for the image enhancement,
with respect to the parameters , , and
when we optimize
. If
is the desired fuzzy contrast, we can find the parameand
by
ters ,
(A26)
The differentiation of
the following:
(A27)
(A28)
(A29)
ACKNOWLEDGMENT
The authors would like to thank the Director, Scientific
Analysis Group, Defence R&D Organization, Metcalfe House,
Delhi, for allowing them to carry out this work.
REFERENCES
[1] J. M. Gauch, Investigation of image contrast space defined by variation of histogram equalization, Graph. Models Image Process., vol.
54, no. 4, pp. 269280, 1992.
[2] P. Mukherjee and B. N. Chatterji, Note: Adaptive neighborhood extended contrast enhancement and its modifications, Graph. Models
Image Process., vol. 57, no. 3, pp. 254265, 1995.
[3] M. Lindenbaum, M. Fischer, and A. Bruckestein, On Gabors contribution to image enhancement, Pattern Recognit., vol. 7, pp. 18, 1994.
[4] L. A. Zadeh, Outline of a new approach to the analysis of complex
systems and decision processes, IEEE Trans. Syst. Man. Cybern., vol.
SMC-3, no. 1, pp. 2844, Jan. 1973.
[5] H. J. Jimmermann, Fuzzy Set Theory and its Applications, Second
ed. Norwell, MA: Kluwer, 1991.
[6] M. Russo and G. Ramponi, A fuzzy operator for the enhancement of
blurred and noisy images, IEEE Trans. Image Process., vol. 4, no. 8,
pp. 11691174, Aug. 1995.
[7] Y. S. Choi and R. Krishnapuram, A robust approach to image enhancement based on fuzzy logic, IEEE Trans. Image Process., vol.
6, no. 6, pp. 808825, Jun. 1997.
[8] W. Niblack, An introduction to Digital Image Processing. Englewood Cliffs, NJ: Prentice-Hall, 1986.
[9] I. M. Bockstein, Color equalization method and its application to color
image processing, J. Opt. Soc. Amer. A, vol. 3, no. 5, pp. 735737,
1986.
[10] R. N. Strickland, C. S. Kim, and W. F. McDonnel, Digital color image
enhancement based on the saturation component, Opt. Eng., vol. 26,
no. 7, pp. 609616, 1987.
[11] M. J. Soha and A. A. Schwartz, Multi-dimensional histogram normalization contrast enhancement, in Proc. 5th Canad. Symp.. Remote
Sensing, 1978, pp. 8693.
[12] S. K. Pal and R. A. King, Image enhancement using smoothing with
fuzzy sets, IEEE Trans. Syst. Man. Cybern., vol. SMC-11, no. 7, pp.
494501, Jul. 1981.
[13] M. Hanmandlu, S. N. Tandon, and A. H. Mir, A new fuzzy logic based
image enhancement, Biomed. Sci. Instrum., vol. 34, pp. 590595,
1997.
[14] H. Li and H. S. Yang, Fast and reliable image enhancement using
fuzzy relaxation technique, IEEE Trans. Syst. Man. Cybern., vol. 19,
no. 5, pp. 12761281, May 1989.
[15] M. Hanmandlu, M. Jha, and R. Sharma, Color image enhancement
using fuzzification, Pattern Recognit. Lett., vol. 24, no. 13, pp. 8187,
2003.
[16] R. C. Gonzalez and R. E. Woods, Digital Image Processing. Reading,
MA: Addison-Wesley, 1992.
[17] F. Russo, An image enhancement technique combining sharpening
and noise reduction, in Proc. IEEE Instrumentation and Measurement
Technology Conf., 2001, pp. 19211924.
[18] S.S. Rao, Optimization: Theory and Applications, 2nd ed. New
Delhi, India: New Age International, 1995.
Madasu Hanmandlu (M02) received the B.E. degree in electrical engineering from Osmania University, Hyderabad, India, in 1973, the M.Tech degree
in power systems from R.E.C. Warangal, Jawaharlal
Nehru Technological University, India, in 1976, and
the Ph.D. degree in control systems from Indian Institute of Technology (IIT), Delhi, in 1981.
From 1979 to 1981, he was a Senior Scientific Officer in the Applied Systems Research Program, Department of Electrical Engineering, IIT Delhi, where
he joined the Electrical Engineering Department as a
Lecturer in 1981 and became a Professor in 1997. He was with the Machine Vision Group, City University, London, U.K., in 1988, and the Robotics Research
Group, Oxford University, Oxford, U.K., in 1993, as part of the Indo-U.K. research collaboration. He was a Visiting Professor with the Faculty of Engineering, Multimedia University, Malaysia, from March 2001 to March 2003. He
worked in the areas of power systems, control, robotics, and computer vision,
before shifting to fuzzy theory. His current research interests mainly include
fuzzy modeling of dynamic systems and applications of fuzzy logic to image
processing, document processing, bio-medical imaging, and intelligent control.
He has authored a book on computer graphics and also has over 160 publications to his credit.
Dr. Hanmandlu is an Associate Editor of the Pattern Recognition Journal,
as well as a Reviewer for several journals, including the IEEE TRANSACTIONS
ON FUZZY SYSTEMS, the IEEE TRANSACTIONS ON IMAGE PROCESSING, and the
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS. He is listed in
Reference Asia, Asias Whos Who of Men and Women of Achievement, 5000
Personalities of the World (1998), and the American Biographical Institute.