2014 Sixth International Conference on Computational Intelligence and Communication Networks

Image Contrast Enhancement using GA and PSO: A Survey

Ravindra Pal Singh, Manish Dixit Sanjay Silakari

Dept. of Comp. Science & IT Dept.Of Computer Science
MITS Gwalior 474005 India UIT ,RGPV Bhopal
[email protected], [email protected] [email protected]

Abstract: Image enhancement as an optimization process is which are seen difficult, or changing the histogram of images
being used rapidly in Image processing system. To enhance the so that they can be interpreted more easily by human
contrast of the images various common methods are proposed observer. The most common techniques used for contrast
in literature including HE and linear Stretching but they give enhancement are linear stretching and histogram equalization
unnatural contrast and brightness in the images.GA as well as and the resultant images from these methods have bright and
PSO both as an optimization method improves contrast and unnatural contrast. It is the fastest growing technology due to
gives better result. This paper reviews an introduction of its various applications in several fields like consumer
several approaches for contrast enhancement in spatial domain electronic products, robotics and medical etc. Enhanced
as well as in frequency domain using Genetic algorithms and
images without human intervention, is very difficult task in
Particle swarm optimization algorithm. The survey of this
topic leads to the conclusion that field of the PSO applications
image processing. This is because automatic enhancement
is growing fast and improving the contrast using PSO will help requires specifying an objective criterion for enhancement
to solve many complex problems in image processing in the while evaluating the quality of an image is done finally by
future. the human interpreter. Most of the enhancement techniques
existent to date are empirical methods, dependent on the
Index Terms: Digital image processing, Image enhancement, particular type of image. More important, these techniques
Histogram equalization, Genetic algorithm, PSO. require interactive procedures to obtain satisfactory results,
I. INTRODUCTION and therefore are not suitable for routine application.
Evolutionary algorithms have been applied to image
Image processing is a very active area of research in enhancement by several authors. The authors apply a global
computing environment. It has many applications in contrast enhancement technique by using Genetic
everyday life tasks, from medicine to transportation and Programming to adapt the color map in the image as to fit the
industry. Producing digital images with hidden details and demands of the human interpreter. The GP was meant to
better contrast is the strong requirement in several areas like partly reproduce the user behavior in interpreting the quality
remote sensing, biomedical image analysis, and fault of the image, but results reported indicated an unsuccessful
detection. For this purpose the methods that process a given behavior. In [10] [8] [9] a global method for image
image so that the result is visually more suitable than the enhancement is used, but it fails to adopt the local
original are called image enhancement techniques. Image information present in the image. A global enhancement
enhancement is an important technique of image processing, method, in which the pixels in the image are modified
refers to highlight some key information in an image and to through a transformation function based on the gray-level
remove some secondary information which aims to improve distribution over an entire image, often doesn’t produce
the quality of identification in the process at same time [1]. satisfying results. Requirement for user interaction, as each
The objective of image enhancement is to make the image, treated as an individual in the population, should be
processed images more suitable for the particular application rated subjectively by a human interpreter.[8] tells that the
than the original image. Thus the results are more in line requirement for additional external parameters in the
with the characteristics of human visual recognition system. objective fitness criterion that make the automatic image
Generally, we can say that enhancement is used to improve enhancement technique parameter dependant. Therefore,
the appearance of an image to make it easier for visual methods based on local enhancement model came in
perception, understanding and analysis of image. There exist literature such as Differential evolutionary algorithm, genetic
two basic categories of approaches to image enhancement: algorithm and particle swarm optimization. This paper
filtering techniques and contrast enhancement methods. In studies about the various existing methods in the literature
the filtering process every pixel values is substituted by a based on the local and global image contrast enhancement in
new value which is calculated through the original value and both spatial and frequency domain. And the overall analysis
its neighborhood pixels value. Contrast enhancement plays is done on the genetic evolutionary algorithm and particle
an important role in image processing system and in this swarm optimization algorithm.
gray level levels of the input image are mapped to a new set
of values in such a way that the histogram of the image II. LOCAL ENHANCEMENT
becomes flatter to get more homogenous distribution of these Local contrast enhancement increases the gray level of
gray levels[2][3]. It is used for expanding the range of
original image on the bases of light and dark edges. The
intensities in a gray scale image, revealing parts of image
local mean and local standard deviation of entire image,

978-1-4799-6929-6/14 $31.00 © 2014 IEEE 186

DOI 10.1109/CICN.2014.51
10.1109/.51
minimum and maximum value of the image are used to successfully explored by Holland in 1960s,after that it
statistically characterize digital image. In local enhancement become more popular[5]. It works well in global solution
the transformation functions are based on the gray-level space. GA is used where exhaustive search for solution is
distribution in the neighborhood of each pixel in the input expensive in terms of computation time. Applications of
image. The following equation (1) is performed on every GAs for image processing extend from evolving filters or
pixel at the location (i, j) in the input image. detecting edges to making complex decisions or classifying
detected features [7]. The simple GA includes 5 steps: First
(1) start with any randomly generated population, after that
Here M is the global mean of the image, m (i, j) and σ (i, compute the fitness value for every chromosomes that is
j) are the local mean and standard deviation respectively, that present in the population. After that selection operator is
are calculated in a neighborhood centered at location (i, performed to select a pair of chromosomes from the
j).That means they are dependent on the local information. population with the help of fitness value and then crossover
f(i , j) and g(i, j) are the gray levels in the input and and mutation operators are applied to produce an offspring.
processed image respectively. The four parameters A, b, c, The whole procedure is repeated until the required offspring
and d are represented as individuals and each individual are created, now if the stopping condition is satisfies the
represents a candidate solution to find out the optimal algorithm stops otherwise it continues. The Crossover
solution for the given problem. These parameters are defined operates by selecting a crossover point and concatenating the
over a real positive numbers and they are same for the whole initial segment of one parent with the final segment of the
second parent to form a new child and second child is
image. . The range of four parameters is given: A ϵ [0.5,
simultaneously generated using the remaining segments of
1.5], bϵ [0, 0.5], c ϵ [0, 1], dϵ [0, 1.5] the range is same as the two parents. Mutation provides for occasional
[12]. The task of optimization algorithm is to solve the image disturbances in the crossover operation by inverting one or
enhancement problem by tuning the four parameters in order more genetic elements during reproduction.
to find the best combination according to an objective
criterion that describes the contrast in the image.
1. Start with randomly generated population of N
III. FITNESS FUNCTION chromosomes.
Before To evaluate the quality of an enhanced image
without human intervention, we need an objective function 2. CalculateTABLE
the fitness
I. value
TABLE for
TYPEeach chromosome in
STYLES
which will say all about the image quality. The objective the population.
Table Table Column Head
function is formed by combining three performance Head Table column subhead Subhead Subhead
measures, namely entropy value, sum of edge intensities and 3. Repeat until N offsprings are created.
copy More table copya
number of edgels (edge pixels). It is observed that compared a). Probabilistically select a pair of chromosomes from
a. Sample of a Table footnote. (Table footnote)
to the original image good contrast enhanced image has current population using the fitness value.
more number of edgels [1] and enhanced version should b). Produce an offspring using crossover and mutation
have a higher intensity of the edges [8]. But these two are Figure 1. Example of a ONE-COLUMN figure caption.
operators.
not sufficient to test an enhanced image, so entropy value is
Please see last page of this document for AN EXAMPLE
added to fitness function to determine the fine details of the
of4.a 2-COLUMN
Replace current population using with newly created
Figure.
image. The following fitness function is considered [13] one.
Figure Labels: Use 8 point Times New Roman for Figure
[17] - labels. Use words rather than symbols or abbreviations when
(2) Fig: Pseudo code for Genetic Algorithm
F(S) is the fitness function. I(S) is the transformed image A real-coded GA with Gaussian uniform crossover is
denotes the input image with applying the transformation proposed, which gives better visualize image compared to
function according to equation (1).The parameters A, b, c GA with arithmetic crossover [9].Real-coded GA with PCA
and d are given by the individuals/particles S=(A b c d). E (principle component analysis) mutation is proposed in [26]
(I(S)) is the sum of M*N pixels intensity of the edges [24]. Analysis of image enhancement based on real-coded
detected by the Sobel edge detector which is applied on the GA, population based incremental learning and
transformed image; here we are using the Sobel operator as comprehensive approach for contrast enhancement is done in
the automatic threshold detector. n_edgels (I(S)) is the [25].An image enhancement method with GA based on
number of the edgel pixels detected through the Sobel edge vector filtering framework is proposed [27] in literature. In
detector. And at last H (I(S)) is the measure of the entropy [28] RGB image is converted into LHS (luminance, hue and
of the image I(S). saturation) image then GA is performed, after that an optimal
set of generalized transform having the largest fitness value
IV. GENETIC ALGORITHM
is determined and as a result LHS image is converted into
A GA is a heuristic search/optimization technique, RGB.
inspired by the natural selection for obtaining the best
possible solution in a huge solution space.GA was first

187
 V. PARTICLE SWARM OPTIMIZATION details in the image. When compared to standard PSO, this
PSO is one of the modern heuristic algorithms that can be method achieves better computation time, efficiency and
maximizes the number of pixel in the image. As the velocity
applied to non linear and non continuous optimization
and particle updation is done only by the Gbest value. In [16]
problems. It is a population-based stochastic optimization [19] multi-objective particle swarm optimization and
technique for continuous nonlinear functions [6]. PSO was modified MPSO is presented for improving the contrast in
developed in 1995 by Dr. James Kennedy and Dr. Russell images as well as to preserve the mean intensity of the
Eberhart.PSO is a nature inspired algorithm that is based on images Gamma-correction method is used [18].
the swarming behavior of the bird flocking and fish
schooling. In PSO the individual are known as particle and Start
the group of the particles is known as swarm. In PSO
particles, are flown through hyper dimensional search space.
Changes to the position of particles within the search space
are based on the social-psychological tendency of Initialize population with random position
Individuals to emulate the success of other individuals [4]. and velocity.
The changes to a particle within the swarm are therefore
influenced by the experience, or knowledge, of its
neighbors. The search behavior of a particle is thus affected For each position of particle, calculate
by that of other particles within the swarm.PSO uses a fitness value
number of particles that constitute a swarm moving around
in the search space looking for the best solution. Each If fitness value is better than Pbest, set
particle in search space adjusts its “flying” according to its current value as new Pbest.
own flying experience as well as the flying experience of
other particles. Each particle adjusts its travelling speed
dynamically corresponding to the flying experiences of Set the best value of Pbest as Gbest.
itself and its neighbors. Each particle modifies its position
according to: its current position, its current velocity, the
distance between its current position and Pbest, the distance Calculate particle velocity and update
between its current position and gbest. The particle’s fitness particle position using equation (3) and (4).
value is evaluated on each iteration. If it is the best value the
particle, the particle stores the location of that value as Pbest
(particle best). The location of the best fitness value
achieved by any particle during any iteration is stored as No
Reach maximum
Gbest (global best).The neighborhood concept in PSO is not
iteration/minimum
the same as in other meta-heuristics search, since in PSO
error
each particle’s neighborhood never changes that means it is
fixed [11]. Yes
Vi=wVi-1 + C1rand (Pbest-pl)+C2rand(Gbest-pl) (3)
Stop
and
pl = pvl + Vi (4)
Fig: Flow-chart for Standard PSO Algorithm
Vi is the current velocity, Vi-1 is the previous velocity, w
Hybrid approach to image enhancement with simulated
is the inertia weight, C1 and C2 are the learning factor, pl is
annealing and PSO, in which the transformation function
the present location of the particle, and pvl is the previous
uses both local and global information of the image. In [20]
position of the particle. It is said that for C1 +C2 ≥4, gives the
LCM-CLAHE based PSO method is proposed which is used
better result, rand is the random number between (0, 1). In
to enhance the contrast for the mammogram images. This
Eqn. (3), the first part represents the inertia velocity of the
method is claimed to give the contrast enhancement by
particle, the second part indicates the particle’s decision
preserving local information in mammogram. In [21]
made by its own experience, and the third part signifies the
Automatic color image enhancement technique based on
swarm’s experience-sociality [4]. In [14] RGB image is
PSO is presented, in this RGB image is converted into HSI
separated into Red, Blue and Green channels, after this PSO
color space and then Using the ‘I’ of HSI intensity images is
is applied to each channel separately then normal HE is
obtained. And now PSO is applied to this intensity image,
applied and at last RGB image is merged back. This
the transformation function uses both local and global
proposed method helps to flattens and stretches the
information. Color RGB image is converted into HSV format
histogram of the images and balances different color
then it is fuzzified using GINT operator and then PSO is
components and it also preserves the brightness of the image.
applied after this defuzzification is performed and the
In [15] a customized PSO is used to improve the contrast and

188
resultant image is converted into RGB image [22]. Wavelet 10. Apurba Gorai and Ashish Ghosh, “Gray-level Image Enhancement By
transform based PSO method is proposed in literature to Particle Swarm Optimization”, IEEE 2009.
determine the high frequency component in the image to 11. C. Munteanu, A. Rosa, “Gray-scale enhancement as an automatic
process driven by evolution,” IEEE Transaction on Systems,Man
work well on low resolution images [23]. andCybernatics-Part B:Cybernetics, vol. 34, no. 2, pp. 1292-1298,
2004.
VI. CONCLUSION
12. Malik Braik, Alaa. Sheta, and Aladdin Ayesh, “Image enhancement
In comparison to linear transformation and histogram using particle swarm optimization”, in Proc. of the World Congress on
equalization, GA treats the image globally and produces Engineering, 2007.
image with natural contrast. The optimization success 13. S. M. W. Masra, P. K. Pang, M. S. Muhammad, and K.
Kipli,“Application of Particle Swarm Optimization in Histogram
depends on the chosen chromosome structure, crossover and Equalization for Image Enhancement”;2012 IEEE Colloquium on
mutation operators as well as the objective function. PSO Humanities, Science & Engineering Research (CHUSER 2012),
and GA are both population based and stochastic December 3-4, 2012, Kota Kinabalu,Sabah, Malaysia 294.
optimization, both algorithms start with a group of a 14. K.Venkatalakshmi and S.Mercy Shalinie, “ A Customized Particle
Swarm Optimization Algorithm for Image Enhancement”,IEEE 2010.
randomly generated population, both have fitness values to
15. Ngai M. Kwok, Q. P. Ha, Dikai Liu and Gu Fang “ Contrast
evaluate the population, both update the population and Enhancement and Intensity Preservation for Gray-Level Images Using
search for the optimium with random techniques and both Multiobjective Particle Swarm Optimization”;IEEE Transactions On
do not guarantee success. PSO does not have genetic Automation Science And Engineering, Vol. 6, No. 1, January 2009
operators like crossover and mutation. Particles update 16. Gora. Apurba, and Ghosh. Ashish, “Gray-level Image Enhancement
themselves with the internal velocity,they also have their By Particle Swarm Optimization”, World Congress On Nature
&Biologically Inspired Computing (Nabic), pp. 72-77, 2009
own memory. The information sharing mechanism in PSO
17. Gao. Qinqing. Zeng. Guangping, Chen, Dexin. He, and Ketai, “Image
is significantly different. There is no selection in PSO: all Enhancement Technique Based On Improved PSO Algorithm”, 6 th
particles survive for the length of the run, PSO is the only IEEE Conference on Industrial Electronics and Applications (ICIEA),
EA that does not remove candidate population members. pp. 234 – 238, 2011.
Since GA uses selection, crossover and mutation operators, 18. N. Kwok, Q. Ha, D. Liu, and G. Fang, “Intensity-Preserving Contrast
Enhancement for Gray-Level Images using Multi-objective Particle
it has a large computation time and is much costly. In GA a Swarm Optimization”, CASE06. IEEE International Conference on
large number of parameters are adjusted during its run, but Automation Science and Engineering CASE-2006, pp. 21-26, 2006
in PSO only a few parameters are adjusted compared to GA. 19. Shelda Mohan and T.R. Mahesh, “Particle Swarm Optimization Based
Therefore, as compared to GA, PSO is simple in concept, Contrast Limited Enhancement for Mammogram Images”IEEE 2013.
easy to implement, cheaper and having small computational 20. Apurba Gorai and Ashish Ghosh, “Hue-Preserving Color Image
time. Enhancement Using Particle Swarm Optimization” IEEE 2011.
21. M. Hanmadlu,Shaveta Arora,Gaurav Gupta and Latika Singh, “A
REFERENCES Novel Optimal Fuzzy Color Image Enhancement Using Particle
Swarm Optimization” IEEE 2013.
1. Rafael C. Gonzalez, and Richard E. Woods “Digital Image
Processing” Second Edition Prentice Hall.2002. 22. Md Iqbal Quraishi, Krishna Gopal Dhal, J Paul Choudhury, Kamal
Pattanayak and Mallika De, “A Novel Hybrid Approach to Enhance
2. S. Jayaraman, S Esakkirajan, and T Veerakumar “Digital Image Low Resolution Images Using Particle Swarm Optimization” 2012 2nd
Processing”. IEEE International Conference on Parallel, Distributed and Grid
3. Antonella Carbonaro , Primo Zingaretti “A Comprehensive Approach Computing.
to Image-Contrast Enhancement”. 23. C. Munteanu and A. Rosa, “Towards Automatic Image Enhancement
4. Andries P. Engelbrecht, “Computational Intelligence: An Using Genetic Algorithms”IEEE 2000.
introduction”,2nd edition. john wailley & sons ltd,2007 24. Akhilesh Verma and Archana, “A Survey on Image Contrast
5. J. Kennedy, R. C. Eberhart, and Y.Shi, Swarm Intelligence.Morgan Enhancement Using Genetic Algorithm”; International Journal of
Kaufmann Publishers, San Francisco,2001. Scientific and Research Publications, Volume 2, Issue 7, July 2012 1
6. Mantas Paulinas, Andrius Ušinskas, “A Survey of Genetic Algorithms ISSN 2250-3153.
Applications for Image Enhancement and Segmentation”, ISSN 1392 – 25. Archana, Akhilesh Verma, Dr. Savita Goel and Neeraj Kumar, “Gray
124x Information Technology And Control, 2007, Vol.36, No.3. Level Enhancement to Emphasize Less Dynamic Region within Image
7. F. Saitoh, “Image contrast enhancement using genetic algorithm,” Using Genetic Algorithm”IEEE 2012.
Proceeding of IEEE SMC, Tokyo, Japan, pp. 899-904, 1999. 26. R. Lukac, K.N. Plataniotis, B.Smolka and A.N.
8. Munteanu C. and Lazarescu V. (1999) “Evolutionary Contrast Venetsanopoulos,“Color Image Filtering and Enhancement Based On
Stretching and Detail Enhancement of Satellite Images”,Proceedings Genetic Algorithms” IEEE 2004
of Mendel’99, Bmo, Czech Rep., pp 94-99 27. Ming-Suen Shyu and Jin-Jang Leou, “A Genetic Algorithm Approach
9. Poli R. and Cagnoni S. (1997) “Evolution of Pseudocolouring to Color Image Enhancement” Elsevier Science Ltd,Pattern
Algorithms for Image Enhancement”. Technical Report: CSRP-975, Recognition, Vol. 31, No. 7, pp. 871-880, 1998.
univ. of Birmingham.

189