ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans

Cuckoo search algorithm based satellite image contrast and brightness

enhancement using DWT–SVD
A.K. Bhandari a, V. Soni a, A. Kumar a,n, G.K. Singh b
a
PDPM Indian Institute of Information Technology Design and Manufacturing, Jabalpur 482011, MP, India
b
Department of Electrical Engineering, Indian Institute Technology Roorkee, Uttrakhand 247667, India

ar t ic l e i nf o a b s t r a c t

Article history: This paper presents a new contrast enhancement approach which is based on Cuckoo Search (CS)
Received 11 November 2013 algorithm and DWT-SVD for quality improvement of the low contrast satellite images. The input image is
Received in revised form decomposed into the four frequency subbands through Discrete Wavelet Transform (DWT), and CS
13 January 2014
algorithm used to optimize each subband of DWT and then obtains the singular value matrix of the low-
Accepted 28 April 2014
low thresholded subband image and finally, it reconstructs the enhanced image by applying IDWT. The
This paper was recommended for
publication by Prof. Y. Chen. singular value matrix employed intensity information of the particular image, and any modification in
the singular values changes the intensity of the given image. The experimental results show superiority
of the proposed method performance in terms of PSNR, MSE, Mean and Standard Deviation over
Keywords: conventional and state-of-the-art techniques.
Adaptive learning
& 2014 ISA. Published by Elsevier Ltd. All rights reserved.
CS algorithm
Wavelet thresholding
DWT–SVD
Image equalization and remote sensing
image contrast enhancement

1. Introduction spectral separation, and because of this property it is feasible to

boost the features by treating singular frequency components in a
Remotely sensed satellite and aerial images are useful in different way [10]. However, there are few drawbacks in trans-
several applications such as in the field of agriculture, geology, forming the remote sensing images using block DCT. Therefore,
forestry, biodiversity conservation, regional planning, education, DWT has also found application in contrast enhancement of
intelligence, warfare, weather forecast, electric power system, etc. images [12].
[1,2]. Images can be in visible colors and in other spectra. In Generally, data sets collected by image sensors are most
general, raw satellite images have a relatively narrow range of commonly contaminated by noise. Various kind of noises like
brightness values; hence, contrast enhancement is frequently used additive noise (Gaussian noise), multiplicative noise (speckle
to enhance the multiband satellite images for better interpretation noise) and artifacts in different imaging modalities corrupt the
and visualization [3,4]. The multispectral remote sensing images images. Imperfect instruments, problems with the data acquisition
are very efficient for better understanding of the earth environ- process, and interfering natural phenomena can all degrade the
ment [3,5]. There have been several techniques to overcome this data of interest. Due to such degradations, it affects human
issue [6–8], such as general histogram equalization (GHE) [6] and interpretation as well as accuracy of the processing methods in
local histogram equalization (LHE) [7]. case of satellite imaging. Computer-aided analysis and quantitative
In general, satellite images contain very low dynamic region of measurements become difficult and unreliable due to poor image
intensity values because of insufficient enlightenment. As a result, quality.
it requires to be processed for better display. In spatial domain, Thus, enhancement of the images becomes necessary for many
numerous methodologies have been developed for the enhance- practical applications. Although, image enhancement is still a
ment of gray level satellite images. These procedures consider challenge for researchers because noise removal introduces arti-
GHE based equalization, gamma correction, high pass filtering facts and it causes blurring of the image. Due to that sometimes a
(HPF), low pass filtering (LPF), homomorphic filtering, etc. [8,9]. In very drastic reduction in information can occur, and over bright-
particular, discrete cosine transform (DCT) domain provides the ness may also take place. The image enhancement research can
be broadly divided into two domains; spatial and transform
domain [8]. Among these two domains, in last two decades, more
n
Corresponding author. investigation has been conducted in wavelet transform (WT)

http://dx.doi.org/10.1016/j.isatra.2014.04.007
0019-0578/& 2014 ISA. Published by Elsevier Ltd. All rights reserved.

Please cite this article as: Bhandari AK, et al. Cuckoo search algorithm based satellite image contrast and brightness enhancement using
DWT–SVD. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2014.04.007i
2 A.K. Bhandari et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

domain. Wavelet based optimization technique for image enhance- contains intensity information of the particular image. In these
ment is very useful and essential research field. Because of having recent works [4,41,42], SVD was used to obtain the ratio of the
inherent qualities of wavelet coefficients such as sparsity and highest singular value of the created normalized matrix, having
decomposition [13], simplified, effective and efficient realization mean zero and variance of one, over a normalized input image that
of the thresholding ideas has become simple. can be accomplished by Eq. (2). In this paper, a new approach has
The ‘wavelet shrinkages’ technique has been introduced by been used for the ratio of largest singular value using Eq. (5). The
Donoho et al. [14,15] for analyzing the wavelet coefficients of real- new ratio approach of SVD is used to preserve maximum edges
world noisy signal, that have been again modified to improve SNR. information so that the enhanced image looks better with respect
The research [16–18] was carried out for improving the perfor- to contrast, brightness and edges.
mance of thresholding, based on these domain. There has been This paper presents an improved contrast enhancement method
continuos focus on improving and excuting the thresholding for low resolution remote sensing images using DWT–SVD and the
function. The Bayes Shrink and Sure Shrink gives good perfor- cuckoo search algorithm. The strength of the paper comes from the
mance which was studied by Foddr and Kamath [19], on the evolutionary CS algorithm, which is a stochastic global optimization
denoising by wavelet shrinkage like soft, hard, garrote and technique that is used for learning the parameters of adaptive
semisoft. Statistical approach like Bayesian approach for denoising thresholding function. The improved thresholding in wavelet
the images has been proposed by the researchers in [20–22], by domain improves overall performance of the state-of-the-art DWT–
experimenting many noise models for the distribution of noisy SVD technique. The proposed technique may play and contribute
wavelet coefficients such as hidden Markov models [23], Gaussian major role in the remote sensing research.
[24], Rayleigh [25] and Fisher-Tippet [26]. This approach has The rest of the paper is organized as follows. Section 2 gives an
been extended by the researchers in [23–29]. These methods are overview of DWT, SVD and optimization scheme. Section 3
dependent on a specific noise model and thus reduce their describes the proposed CS algorithm based satellite image contrast
flexibility. Further, some of the researchers [28–32] have proposed and brightness enhancement using the DWT–SVD domain. Section
a mixture of statistical models, which is more computationally 4 discusses the qualitative and quantitative results of the proposed
complex. method along with the conventional and state-of-the-art contrast
The computational complexity has been increased by using and brightness enhancement techniques supported by Mean,
spatial operation. Further, the improvement is gained by proper Standard Deviation, PSNR and MSE. Conclusions are given in the
shrinkages in an image using an optimum threshold value com- final section.
puted in sub-band adaptive method based on either wavelet
transform or wavelet packets or DCT [33–38]. Bhutada et al. have
proposed a Particle Swarm Optimization (PSO) based learning of 2. Overview of DWT and SVD
sub-band adaptive thresholding function for image denoising [39].
Nasri and Pour [36] proposed a wavelet domain based denois- DWT [41–44] is one of the recent wavelet transforms used in
ing technique using a new adaptive thresholding function termed image processing. Wavelet analysis is basically about the breaking
as wavelet transform based thresholding neural network (WT- up of a signal into shifted and scaled versions of the original (or
TNN) method in 2009. They have reported that their methodology mother) wavelet.
outperforms the various other thresholding methodologies such
as soft, hard, garrote and other existing WT-TNN methodologies.
2.1. Discrete wavelet transform (DWT)
Furthermore, there were various limitation in the WT-TNN
approach. Proper initialization of threshold value and other
Discrete wavelet transform has emerged as a dominant and
thresholding parameters is required in this methodology; other-
effective mathematical tool in the field of remote sensing and
wise it imposes a limitation on the speed of convergence of
image processing. The basic concept of DWT is decomposing the
learning process. The limitations of the WT-TNN approach have
given input signal into four parts using the property of translation
been overcome by the PSO algorithm in place of steepest descent
and dilation, which is known as mother wavelet. The appropriate
gradient-based LMS techniques. Researchers [40] have proposed
wavelet function is to be selected for wavelet decomposition of the
the image enhancement by WT-TNN with adaptive learning rate.
image for obtaining wavelet coefficients [41]. A systematic block
This paper introduces a new method that is based on the use
diagram of DWT filter banks of level 1 is shown in Fig. 1.
of DWT, wavelet thresholding, optimizing the thresholds using
The 2-D DWT decomposition of any given input image can be
a specific optimization strategy (CS), and the use of SVD for
obtained by employing 1-D wavelet transform along the rows
enhancement. Therefore, with the aim of providing contrast
of the image first, and then the resultant are decomposed along
enhancement and detail emphasis of satellite images, a new
the columns. This action splits the given input image into four
CS–DWT–SVD algorithm has been proposed with a new approach
decomposed subband images, which are known as LL, LH, HL and
of optimization technique using thresholding function, which is a
HH frequency bands. This routine can be processed to obtain
modified and extended version of the previous works [4,41,42]. In
multiple “scale” wavelet decomposition. DWT decomposes any
order to indentify the performance of the DWT–SVD technique, a
verity of wavelet filters for low contrast and brightness enhance-
ment of satellite images has been proposed [41], where Meyer Lowpass 2 LL
wavelet and SVD based results showed better performance than
the other wavelet filters. In 2011, a new satellite image contrast Lowpass 2
Highpass 2 LH
enhancement technique based on SVD–DCT has been proposed
[4,42] and it is reported that the basic enhancement occurs due to Input
scaling of singular values of the DCT coefficients. Image Lowpass
2 HL
DWT–SVD and DCT–SVD based image illumination enhance-
ment technique presented in [41,42], is based on updating the Highpass
2
singular value matrix obtained by SVD in low–low subband of Highpass 2 HH
input image by applying DWT. The main motivation behind using
SVD in [41,42] for image enhancement comes from the fact that ΣA Fig. 1. Block diagram of DWT filter banks of level 1.

Please cite this article as: Bhandari AK, et al. Cuckoo search algorithm based satellite image contrast and brightness enhancement using
DWT–SVD. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2014.04.007i
 A.K. Bhandari et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3

Fig. 2. (a) Sine wave and (b) wave of wavelet db10.

given signal into a set of basic functions, which is known as decomposition (SVD) of A and can be written as
wavelets. It is realized from a single prototype wavelet ψ(t) by
mother wavelet using dilations and shifting [8]. The main advan- A ¼ U A ∑A V TA ð1Þ
tage of DWT is that it gives more accurate and efficient results as
compared to the FFT [45] and DCT [4]. where, UA and VA are orthogonal square matrices called as hanger
A wavelet is a waveform which denotes an effectively limited and aligner correspondingly.
duration with an average value of zero. The frequency components ΣA matrix included the array of singular values on its main
of DWT decomposed subbands used to deal with full frequency diagonal, and primary enhancement occurs due to scaling
spectrum of the given image signal. One most important advan- of singular values of DWT coefficients. The singular value
tage existing in the wavelets is the capability to perform local matrix represents the intensity information of image, and any
analysis that is used to analyze a localized area of a larger signal. alteration on singular values changes intensity of the input
Fig. 2(a) sine waves and Fig. 2(b) wavelets are indicating intui- image. The main advantage of using SVD for image equalization
tively, that represent signals with sharp changes and might be comes from the fact that ΣA contains intensity information of
better analyzed with an irregular wavelet than with a smooth the image [47,48]. In case of singular value decomposition, ratio
sinusoid. It also makes sense that local features can be described of highest singular value of generated normalized matrix, with
better with wavelets that have local extent. mean zero and variance of one, for a particular image can be
computed by
2.2. Singular value decomposition (SVD)

max ∑Nðμ ¼ 0;var ¼ 1Þ
ξ¼ ð2Þ
In linear algebra, SVD is a decomposition of a real or complex maxð∑A Þ
matrix, with numerous helpful properties in image signal proces-
sing and other filed. It is a numerical technique used to diagonalize where, ΣN(μ ¼ 0,var ¼ 1) is the singular value matrix of the synthetic
matrices. SVD can be calculated mainly by three mutually compa- intensity matrix. These coefficients can be used to regenerate an
tible terms. In other words, SVD is a method for decomposing equalized image using
correlated variables into a set of uncorrelated ones which is
used to give improved representation of numerous relationships EequalizedA ¼ U A ðξ∑A ÞV TA ð3Þ
between the original data items [4,41,46]. Simultaneously, it is also
a process of recognizing and organizing the dimensions along where, Eequalized A is used to denote the equalized image, termed as A.
which data points exhibit the most variation [47]. Therefore, SVD Singular values provide energy information of the image as
[47,48] is known as a technique for data reduction and typically for well as knowledge of how the energy is distributed [52–54]. For
feature detection and for enhancement purpose [49]. Following example, for an image with random textural content, its energy
are the essential ideas using SVD: considering a high dimensional, will spread over all the singular values (all the singular values are
highly changeable set of data points and decreasing it to a lesser significant). On the other hand, for a smooth image, the first few
dimensional space that exposes substructure of the original data singular values will be dominant, while all others are less signifi-
more smoothly and rearranges it from most variation to the least cant. In most of the cases, the first (largest) singular value roughly
[48–50]. corresponds to mean of the image, thus closely relating to the
The SVD technique represents a rectangular matrix A which can spectral features, while all other singular values provide detailed
be decomposed into the product of three matrices. Every real information about spatial content of the image that relates to the
matrix A can be broken down into a product of three matrices, textural features.
acknowledged as A¼UΣVT, where, U and V employed as orthogo- In this work, DWT is applied on multispectral satellite image of
nal matrices. In the matrix, the diagonal component of Σ identified each band such as NIR, red and green band to improve better
as singular values of matrix A, columns of U identified as left texture features presented in low contrast image. DWT transforms
singular vectors of A, and columns of V are known as right singular original image into four subbands, as LL, LH, HL, HH [41,47]. The
vectors of A [51]. frequency components of these subband images cover all the
Image equalization using the SVD technique depends upon frequency components of the original image. Hence, after IDWT,
equalizing the singular value matrix realized through SVD [52]. enhanced image using the CS algorithm and SVD will be more
The complete decomposition process is called as singular value effective, sharper and having a high-quality contrast.

Please cite this article as: Bhandari AK, et al. Cuckoo search algorithm based satellite image contrast and brightness enhancement using
DWT–SVD. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2014.04.007i
4 A.K. Bhandari et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 3. Behavior of thresholding function with respect to variation in thresholding parameters (a) Effect of variation of k between (0, 1) with m¼2, λ¼ 5. (b) Effect of variation
of m between (2, 10) with k ¼1, λ ¼5 [36].

2.3. Brief explanation of optimization process

Low Inverse
Contrast Adaptive Thresholding DWT
The basic idea of enhancing of an image using wavelet trans- Function (λ, k, m)
Image
form is as follows: First take wavelet transform of the input image.
Then wavelet coefficient is modified by the suitable thresholding
function. Finally, after the modification of wavelet coefficient, DWT of Optimization Technique Enhanced
inverse wavelet transform is applied to obtain the reconstructed Image Image
image. In wavelet based thresholding technique, thresholding
function has major effect on the quality of image. Therefore, many
Fig. 4. Wavelet domain thresholding [39].
types of thresholding function were introduced having the prop-
erty of hard, soft, semisoft and garrote. The garrote and semisoft
thresholding functions are an improved way of thresholding. thresholding function. Instead of learning weights of classical
These thresholding functions have the property and advantage of networks, learning process focuses on learning threshold value
both hard and soft thresholding. There is no flexibility exhibit as of the thresholding function. It is extended for each sub-band of
these functions are owing to their dependency and fixed structure the wavelet domain as shown in (Fig. 4) by the researchers [36].
on a fixed threshold value. For reducing the above drawback, in In this approach, they have used the adaptive least mean
some papers, several classes of thresholding functions have been square (LMS) algorithm with an adjustment of the learning para-
proposed with several shape tuning parameters that makes it meter with the help of steepest descent gradient of mean square
more flexible in usage. error (MSE) risk or Steins unbiased risk estimation (SURE)
Zhang [34,35] proposed a form of soft thresholding function. depending on availability/unavailability of the original (noise free)
For better flexibility and capability, Nasri and Pour [36] extended image, respectively. Performance of the proposed technique over
Zhang's proposed functions by adding three shape tuning factors performs all the available state-of-art methods for image contrast
that are given in the following equation: and brightness enhancement. The visual and quantitative results
8 are given in the results and discussions section.
Y  0:5Y kmλ  1 þ ðk  1Þλ Yij 4 λ
m
>
>
> ij
> ij
>
< mþ2k

ηNasriðY ij ; λ; k; mÞ ¼ 0:5kjYmij jþ 2  k signðY ij Þ Yij o ¼ λ

k
ð4Þ 3. Proposed methodology
>
> λ k
>
>
>
: Y ij þ 0:5kð mλÞ1  ðk  1Þλ
m
Yij o  λ
Yij
At present one is commonly faced with a situation that appeals
appropriate enhancement. Usually an image holds several ill-defined
where, η is thresholded wavelet coefficient, Yi,j is wavelet coeffi- and ambiguous areas. The gray value allocated to a pixel is also the
cient of input, λ is threshold, λ, k and m are jointly tunning typical reflectance of many kinds of land covers. Therefore, in case of
parameters, sign represents negative or positive sign of Yi,j and satellite images, precisely finding the enhanced image in the existence
0.5, 2.0 are constants. In this paper, range of λ, k, m are selected as of inherent uncertainty and ambiguity is a challenging task.
[1, 0.1, 1] to [150, 1, 4] respectively.
The thresholding function η varies from hard to soft thresh- 3.1. DWT–SVD based enhancement of low contrast satellite images
olding by adjusting the parameter k [0, 1] in the given expression.
The flexibility in adjustment of thresholding from hard to soft The proposed work of enhancement for the satellite image is
because of variation of k is shown in Fig. 3(a). The shape of carried out in two parts. The first one is SVD and is singular value
thresholding function is decided by the parameter m which adds matrix, which is mentioned in Section 2. The SVD contains
more flexibility to the thresholding function as shown in Fig. 3(b). illumination information in the image, so that conversion of the
Symbol λ is threshold value, which has a major role in threshold- singular values will directly change illumination of the image, and
ing operation. other information present in the image will be as same as before.
In thresholding function, the desired optimum value of thresh- As mentioned in Sections 1 and 2, the enlightenment information
old is obtained by optimum value of λ, k and m. To obtain optimum is surrounded in the LL subband. The edges are concentrated in
value parameters, Zhang [34,35] has proposed the TNN metho- other subbands (i.e., LH, HL, and HH).
dolgy based on neural networks concept. In this type of neural Therefore, separating the high-frequency subbands and apply-
network, activation function has been replaced by the wavelet ing the illumination improvement and thresholding using the CS

Please cite this article as: Bhandari AK, et al. Cuckoo search algorithm based satellite image contrast and brightness enhancement using
DWT–SVD. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2014.04.007i
 A.K. Bhandari et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 5

algorithm in LL subband will only protect the edge information After getting optimized DWT components and direct DWT
from possible degradation and then subsequent to reconstructing component, SVD is applied for the calculation of U, Σ, V and then
the last image by using inverse of DWT. The final satellite image the max element in U1 from LL^ and U2 from LL respectively are
will not only be enhanced with respect to illumination but will obtained.
also be sharper. The correction coefficient for the singular value matrix can be
A complete flowchart routine for the proposed method is calculated by using
shown in Fig. 5. maxðU 1Þ
The general method of the proposed technique is as follows. In ξ¼ ð5Þ
maxðU 2 Þ
the proposed technique, initially the low contrast input multi-
spectral satellite image ‘Ai’ is processed by GHE to generate ‘A^i ’. Σ ^1 ¼ ξðΣ 1 Þ ð6Þ
After getting this, both of these images are transformed by
DWT into LL, LH, HL, and HH. The CS algorithm is applied for Inverse SVD ¼ U 1 ðΣ ^1 ÞV 1 T ð7Þ
optimization of each band. Thresholded subbands of DWT are
created such as LL^, LH^, HL^ and HH^. After taking inverse SVD, it will reconstruct new LL^SVD subband
image. Now, the (LL^SVD , LH^, HL^, HH^) subband images of the
original image are recombined by applying IDWT to generate the
resultant optimized equalized satellite image.

Low contrast input colored satellite image Ai ¼ ðLL^SVD ; LH^; HL^; HH^Þ ð8Þ
(Ai) where i € {R, G, NIR} i.e. AR, AG, ANIR
The mechanism of contrast enhancement can be attributed to
thresholding of DWT component and scaling of singular values of
LL^ coefficients. Since, singular values denote luminance of each
Equalized image image layer after decomposition, scaling of these values leads to
using GHE variation (enhancement) of luminance of each layer and hence,
leads to overall contrast enhancement.
Following steps are undertaken to explain the main computa-
DWT of (Ai^) DWT of (Ai)
tional process of the proposed algorithm:

Step 1: In the very first step, a low contrast multispectral

LL LH HL HH LL LH HL HH satellite image has been taken for the processing.
Step 2: Equalize the satellite image using the general histogram
Cuckoo Search Algorithm equalization technique.
for Optimization Step 3: After equalization, compute 1-level DWT decomposi-
tion of the input image for contrast enhancement of each band
such as NIR band, Red band and Green band.
Thresholded subbands of DWT Step 4: The cuckoo search algorithm is applied for optimization
of each subband of DWT.
Step 5: Thresholded subbands of DWT are created such as LL^,
LL^ LH^ HL^ HH^ LH^, HL^ and HH^.
Step 6: After getting optimized DWT components and direct
DWT component, SVD is applied for the calculation of U, Σ, V
SVD, Calculate U1, Σ1, V1 and SVD, Calculate U2, Σ2, V2 and and then max element in U1 from LL^ and U2 from LL respec-
find the max element in U1 find the max element in U2 tively are obtained.
Step 7: Calculate ξ using Eq. (5) ξ ¼ maxðU 21ÞÞ
maxðU

^
Max U1 from LL^ Max U2 from LL Step 8: Compute new Σ^1 using Eq. (6) Σ 1 ¼ ξðΣ 1 Þ
Step 9: After computing new Σ^1 , apply the inverse of SVD
using Eq. (7) Inverse SVD ¼ U 1 ðΣ ^1 ÞV 1 T for generation of new
Calculate ξ using
max (U1 ) LL^SVD .
ξ= Step 10: Apply IDWT using (LL^SVD , LH^, HL^, HH^) after getting
max (U 2 )
new LL^SVD .
Step 11: Optimized enhanced multispectral satellite image.
New Σ1^= ξ(Σ1)

3.2. Cuckoo search algorithm

Inverse of SVD = U1 (Σ1^) V1T
The cuckoo search algorithm is a recently developed meta-
New LL ^ heuristic optimization algorithm, which is suitable for solving
SVD
optimization problems. In 2009 researchers [55] have proposed
the cuckoo search optimization algorithm. It is inspired by the
IDWT using (LL^SVD, LH^, HL^, HH^) obligate brood parasitism of some cuckoo species by laying their
eggs in the nests of other host birds (of other species). Some host
birds can engage direct conflict with the intruding cuckoos. The
Optimized enhanced satellite image
algorithm is based on the obligate brood parasitic behavior of
Fig. 5. Block diagram of the proposed contrast and brightness enhancement some cuckoo species in combination with the L0 evy flight behavior
algorithm. of fruit flies and some of the birds. In order to Lévy flight

Please cite this article as: Bhandari AK, et al. Cuckoo search algorithm based satellite image contrast and brightness enhancement using
DWT–SVD. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2014.04.007i
6 A.K. Bhandari et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

distribution, it is found that animals and birds search for food in a

random or quasi random manner, and essentially follow a random Subbands of
walk because the next step is based on the current place and the DWT
transition probability to the next state. Such behavior has been
applied in cuckoo search optimization by Yang and Deb [55] and
its superiority over other distribution based random walk pro- For each decomposed detail
coefficient matrix
blems has been demonstrated. Each egg in cuckoo search algo-
rithm represents a solution and cuckoo egg represents a new
solution. Overall, there is aim to employ the new and potentially
Initialize the search parameters randomly
better solutions to replace weak solutions in the nests. In the
simplest form, each nest has one egg.
In nature, animals investigate for food in a random or quasi- Threshold using adaptive thresholding
random way. Usually, the foraging path of an animal is effectively a function
random walk because the next move is based on the current
position/state and the transition probability to the next location.
The selected path absolutely depends on a probability, which can Calculate MSE i.e. fitness and keep current
best (minimum)
be modeled mathematically. Various studies have shown that the
flight behavior of many animals and insects demonstrates the
typical characteristics of Lévy flights [56]. A Lévy flight is a random Iteration (i)
move in which the step-lengths are disseminated according to a
heavy-tailed probability distribution. After a large number of steps,
Generate new nest (solutions) by levy
the distance from the origin of random walk tends to a stable
flight (but keep the current best Repeat
distribution [57]. fitness) and update (i) till
Many nature inspired algorithms have been used for optimiza-
tolerance
tion purpose but the reason behind using cuckoo search is due to limit is
the following properties: Discovery, randomization of new nest
obtained
and evaluate fitness for this set
1. CS is a population based algorithm. The entry wise product is
similar to PSO but random walk via Lévy flight is more efficient
in discovering the search space as its step length is much longer Update fitness to get minimum best
in the lengthy execution.
2. The number of parameters tuned are much less than the Threshold using best parameters
genetic algorithm (GA), PSO and artificial bee colony (ABC).
Therefore, it is potentially more suitable for a wide class of the
optimization procedures. Thresholded
subbands of DWT
The cuckoo algorithm can be extended to more complicated
cases, which are having more than one egg representing a set of Fig. 6. Detailed steps of the cuckoo search algorithm for optimization.
solution. The CS is based on three idealized rules:

a. Each cuckoo lays one egg at a time, and dumps it in a randomly equation:
chosen nest; xi ðt þ 1Þ ¼ xðtÞi þ α  L'evyðλÞ ð9Þ
b. The best nests with high quality of eggs (solutions) will carry
over to the next generations; where, α is step size.
c. The number of available host nests is fixed, and a host can It essentially provides a random walk while random step length
discover an alien egg with probability pa A [0 1]. In such case, is drawn from a L'evy distribution, which has an infinite
the host bird either abandons the nest to build a completely variance with an infinite mean. L'evy distribution is given by
new nest in a new location or throw the egg away. L'evy  u ¼ t  λ ; ð1 o λ r 3Þ ð10Þ

Complete flow chart routine of the CS algorithm is shown L'evy function can be changed according to application. Man-
in Fig. 6. tegna's algorithm is one of the L0 evy function.
On following the above mentioned rules, basic steps of cuckoo Step 5: Evaluate this set of solutions and obtain the new fitness.
algorithm are: Compare the old fitness with this new fitness and replace old
fitness if new fitness is better than the old one. Update the best
Step 1: Set the number of nest. Nest is nothing but different nest corresponding to fitness.
solutions. In this problem, it is taken as 20. Set the probability Step 6: Repeat the above process until some stopping criteria is
with a discovery rate (probability). Set the stopping criteria, achieved giving the best fitness and corresponding best nest.
which are either fixed number of iteration or tolerance value.
Set dimension of the problem, number of dimension is 3 and The quality of input image was poor but after applying CS–
also set boundaries of the parameters. DWT–SVD, result is enhanced and optimized with reference to
Step 2: Randomly initialize the solution, by generating n brightness and contrast. The histograms obtained by the proposed
different nest for obtaining n different solutions. technique are stretched in dynamic range, thereby signifying
Step 3: Evaluate fitness for each of the obtained solution. Find improvement in contrast of the output image. The mechanism
best nest corresponding to minimum value of fitness. of contrast enhancement can be attributed to scaling of singular
Step 4: Start iteration, generate new nest by Levy flight values of DWT coefficients. Since singular values denote luminance
but keep the current best. A Levy flight is performed by the of each image layer after decomposition, scaling of these values

Please cite this article as: Bhandari AK, et al. Cuckoo search algorithm based satellite image contrast and brightness enhancement using
DWT–SVD. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2014.04.007i
 A.K. Bhandari et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 7

Fig. 7. (a1–d1 and a2–d2) Low contrast satellite images, (e1–h1 and e2–h2) enhanced images using GHE technique, (i1–l1 and i2–l2) enhanced images using DCT–SVD
technique, (m1–p1 and m2–p2) enhanced images using DWT–SVD technique, and (q1–t1 and q2–t2) enhanced satellite images based on proposed DWT–SVD using the CS
algorithm [58–61].

Please cite this article as: Bhandari AK, et al. Cuckoo search algorithm based satellite image contrast and brightness enhancement using
DWT–SVD. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2014.04.007i
8 A.K. Bhandari et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 7. (continued)

Please cite this article as: Bhandari AK, et al. Cuckoo search algorithm based satellite image contrast and brightness enhancement using
DWT–SVD. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2014.04.007i
 A.K. Bhandari et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 9

leads to variation (enhancement) of luminance of each layer and can be considered as a random variable with a distribution
hence, leads to the overall contrast enhancement. function. At first, performance of the proposed algorithm is carried
out on multispectral satellite images sample. Thereafter, compar-
ison of the proposed CS optimization method is done over GHE
4. Results and discussion DCT–SVD and DWT–SVD which shows superiority of the proposed
technique.
Satellite images often need enhancement in the presence of PSNR block computes the peak signal-to-noise ratio, in decibels
uncertainty, caused due to the factors like highly dependent on between two images. This ratio is often used as a quality
environmental conditions, poor resolution and poor illumination measurement between the original and improved images. Higher
and have very low spatial resolution. Therefore, in this paper, a PSNR signifies better quality of the enhanced or reconstructed
new approach using the CS algorithm with DWT–SVD for multi- image. Mean Square Error (MSE) and Peak Signal to Noise Ratio
spectral images has been introduced for efficient enhancement of (PSNR) are the two error metrics used to compare enhanced
the satellite images. Owing to complexity problem in satellite quality of the image. MSE represents cumulative squared error
image processing, most of the existing enhancement work have between the enhanced and original image, whereas PSNR repre-
been examined in the normal date set or only in the gray level sents a measure of the peak error. Lower value of MSE represents
images. In the satellite image, rate of information is very high lower error.
because of that existing features in the image is very dense due to To compute PSNR, block, at first, calculates the mean-squared
that rate of change from one region to other region is very rapid. error by using the following equation:
Therefore, In case of enhancement of remote sensing images or
satellite images, accurate processing is a very challenging task. ∑M;N ½I 1 ðm; nÞ  I 2 ðm; nÞ2
MSE ðαÞ ¼ ð13Þ
Accordingly, to achieve better enhancement of satellite images, MN
combination of two robust techniques named CS and DWT–SVD
Here, M and N are number of rows and columns in the input images,
are utilized in this paper, which shows the effectiveness of their
respectively, I1 is the original image and I2 is the thresholded image
enhanced result.
or optimized image using cuckoo search algorithm. Then, the block
In this section, performance of GHE, DCT–SVD, DWT–SVD and
computes PSNR using the following equation:
the proposed DWT–SVD using the CS algorithm is evaluated by
considering fidelity of the reconstructed image to original image. !
R2
For this, Mean, Standard Deviation, PSNR and MSE fidelity assess- PSNR ðβ Þ ¼ 10 log 10 ð14Þ
MSE
ment parameters are considered. Different satellite images are
included to demonstrate usefulness of this algorithm. Performance
where, R is the maximum fluctuation in input image data type.
of this method is measured in terms of following significant
There are several methods, which have been used for satellite
parameters:
image contrast and brightness enhancement. In this paper, one
1 M 1 N1 conventional (GHE) technique and 2 state-of-the-art (DCT–SVD
Mean ðμÞ ¼ ∑ ∑ Iðx; yÞ ð11Þ
MN x ¼ 1 y ¼ 1 and DWT–SVD) techniques are used for comparison purposes.
Fig. 7(a1–d1 and a2–d2) show the low contrast satellite images.
rffiffiffiffiffiffiffiffi
1 M 1 N 1 These images have been equalized by using GHE [Fig. 7(e1–h1 and
Standard Deviation ðsÞ ¼ ∑ ∑ fIðx; yÞ  μg2 ð12Þ e2–h2)], DCT–SVD [Fig. 7(i1–l1 and i2–l2)], DWT–SVD [Fig. 7(m1–p1
MN x ¼ 1 y ¼ 1
and m2–p2)] and the proposed equalization technique DWT–SVD
Mean (m) is the average of all intensity value. It denotes average using the CS Algorithm [Fig. 7(q1–t1 and q2–t2)]. Proposed algorithm
brightness of the image, where as standard deviation is the gave good mean value and minimum standard deviation. PSNR
deviation of intensity values about mean. It denotes average and MSE are also indicating comparatively maximum and minimum
contrast of the image. Here, I(x, y) is the intensity value of pixel values for all the 8 sample satellite images. It was also tested for
(x, y), and (M, N) are dimension of the image. Any pixel of an image many more low contrast satellite images.

Table 1
Comparison of the results between input, enhanced GHE, DCT–SVD, DWT–SVD and proposed DWT–SVD using the cuckoo search algorithm.

Sample Input image Output GHE Output DCT–SVD Output DWT–SVD Output of proposed DWT–SVD
images using CS algorithm

Mean (m) and Mean (m) and PSNR (dB) Mean (m) and PSNR (dB) Mean (m) and PSNR (dB) Mean (m) and PSNR (dB)
variance (s) variance (s) and MSE variance (s) and MSE variance (s) and MSE variance (s) and MSE

1 m ¼124.0820 m ¼ 127.563 α ¼27.657 m ¼ 131.3930 α ¼42.2362 m ¼ 128.7648 α ¼43.9727 m ¼ 129.6809 α ¼ 44.8972

s ¼1.473eþ 003 s ¼5.590e þ003 β ¼ 111.55 s ¼ 6.018e þ003 β ¼ 3.8856 s ¼ 4.576eþ 003 β ¼ 2.6050 s ¼ 6.279eþ 03 β ¼ 2.2523
2 m ¼114.8540 m ¼ 127.572 α ¼27.939 m ¼ 143.5694 α ¼41.8017 m ¼ 139.7473 α ¼44.4140 m ¼ 132.4598 α ¼ 45.3982
s ¼937.7782 s ¼5.597eþ 003 β ¼ 104.52 s ¼7.428eþ003 β ¼ 4.2945 s ¼ 5.290e þ003 β ¼ 2.3533 s ¼ 9.288e þ03 β ¼ 2.0031
3 m ¼169.0433 m ¼ 127.570 m ¼ 25.867 m ¼ 145.8880 α ¼42.0715 m ¼ 124.6742 α ¼43.3654 m ¼ 131.2351 α ¼ 44.9646
s ¼1.069eþ 003 s ¼5.598e þ003 s ¼ 168.418 s ¼7.446eþ 003 β ¼ 4.0358 s ¼ 4.384e þ003 β ¼ 2.9960 s ¼ 8.359eþ 03 β ¼ 2.2523
4 m ¼188.4100 m ¼ 128.324 α ¼25.451 m ¼ 109.8149 α ¼42.2018 m ¼ 134.8720 α ¼43.1670 m ¼ 134.4431 α ¼ 45.7896
s ¼1.381e þ003 s ¼5.773eþ 003 β ¼ 185.374 s ¼4.357e þ003 β ¼ 3.9165 s ¼ 4.679eþ 003 β ¼ 3.1361 s ¼ 1.044eþ 03 β ¼ 1.7268
5 m ¼134.3337 m ¼ 127.360 α ¼26.855 m ¼ 151.1095 α ¼42.9643 m ¼ 119.6784 α ¼40.4032 m ¼ 141.2435 α ¼ 47.8914
s ¼1.570eþ003 s ¼5.593e þ003 β ¼ 134.18 s ¼8.961eþ 003 β ¼ 3.2859 s ¼ 4.610e þ003 β ¼ 5.9260 s ¼ 1.008e þ03 β ¼ 1.1173
6 m ¼95.5495 m ¼ 127.559 α ¼30.197 m ¼ 135.9326 α ¼49.5644 m ¼ 133.4787 α ¼49.5058 m ¼ 139.6562 α ¼ 52.5163
s ¼3.245eþ 003 s ¼ 5.603e þ003 β ¼ 62.1498 s ¼6.712eþ 003 β ¼ 0.7188 s ¼ 5.479e þ003 β ¼ 0.7286 s ¼ 8.568eþ 03 β ¼ 0.3994
7 m ¼179.5113 m ¼ 127.364 α ¼25.123 m ¼ 128.0500 α ¼43.0583 m ¼ 119.7880 α ¼40.3139 m ¼ 149.1953 α ¼ 46.5870
s ¼2.434eþ 003 s ¼5.572e þ003 β ¼ 199.901 s ¼5.704e þ003 β ¼ 3.2155 s ¼ 4.764eþ 003 β ¼ 6.0490 s ¼ 7.666eþ 03 β ¼ 1.4269
8 m ¼108.4103 m ¼ 127.506 m ¼ 29.731 m ¼ 126.1237 α ¼43.9801 m ¼ 125.5332 α ¼45.6989 m ¼ 133.8241 α ¼ 48.8721
s ¼1.719e þ003 s ¼5.593e þ003 s ¼ 69.1829 s ¼5.594e þ003 β ¼ 2.6006 s ¼ 4.748eþ 003 β ¼ 1.7506 s ¼ 7.499eþ 03 β ¼ 0.9219

Please cite this article as: Bhandari AK, et al. Cuckoo search algorithm based satellite image contrast and brightness enhancement using
DWT–SVD. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2014.04.007i
10 A.K. Bhandari et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

In order to show effectiveness of the proposed method over 5. Conclusion

conventional and state-of-the-art contrast and brightness image
enhancement techniques, 4 low contrast 3 band INSAT images and In this paper, an improved method based on evolutionary
4 low contrast LANDSAT images with different features were used algorithms for enhancement of satellite images is proposed. This
for comparison of results. Table I compares the Mean, Variance paper presents a novel contrast enhancement method for low
PSNR and MSE performance of the proposed technique with resolution remote sensing images using the DWT–SVD and CS
conventional GHE and state-of-the-art DCT–SVD and DWT–SVD algorithm. The singular value matrix represents the intensity
techniques. information of the given image, and any changes on singular
The visual results of the proposed technique are shown in values change the intensity of the input image. The proposed
Fig. 7(q1–t1 and q2–t2) and supported quantitative results are technique converts the image into DWT–SVD domain and after
shown in Table 1. For example, in Fig. 7(q1) output enhanced normalizing the singular value matrix; the enhanced image is
image, PSNR of proposed technique are 17.2402 dB, 2.661 dB and reconstructed by using IDWT. The proposed technique has been
0.9245 dB higher than the PSNR obtained by GHE, DCT–SVD and tested on several satellite images, where their PSNR and visual
DWT–SVD techniques, respectively whereas MSE of proposed results show superiority of the proposed technique over conven-
technique are 109.2977 and 1.6333, 0.3527 lesser than the MSE tional and state-of-the-art image contrast and brightness enhance-
obtained by using the GHE, DCT–SVD and DWT–SVD techniques, ment techniques. The main reason of the contrast and brightness
respectively. Results in Table 1 clearly indicate superiority of the enhancement is the DWT–SVD and optimization technique i.e. the
proposed technique over other state-of-the-art and conventional cuckoo search algorithm. The basic enhancement occurs due to
image contrast and brightness enhancement techniques. Charac- scaling of singular values of DWT coefficients and it is optimized
terization of the quantitative performance using metrics denoting by the CS algorithm.
contrast enhancement, color enhancement and perceptual quality Visual examination of the enhancement results reveals that the
shows noteworthy performance of the proposed technique for proposed algorithm effectively distinguishes details with great
satellite images. overall contrast enhancement. The algorithm is expected to be
In this paper, CS algorithms with the DWT–SVD method are useful for various remote sensing applications and also significant
employed. This method is considered for satellite images and to detect surface features of the visible area, which are especially
remotely sensed images are having some band combination such advantageous for the municipal planning and management.
as in case of INSAT image it is having 3 band data, and in case of
LANDSAT images it is having 7 band data. In this study, four INSAT References
and four LANDAT images with different features are used for better
comparison of the results. Generally, any kind of practical proces- [1] Bhandari AK, Singh VK, Kumar A, Singh GK. Cuckoo search algorithm and wind
sing of satellite images has been performed in the three bands driven optimization based study of satellite image segmentation for multilevel
named NIR, Red and Green bands. The specification of all the band thresholding using Kapur's entropy. Expert Syst Appl 2013;41(7):3538–60.
[2] Soni V, Bhandari AK, Kumar A, Singh GK. Improved sub-band adaptive
combination is given in Table 2. thresholding function for denoising of satellite image based on evolutionary
Owing to the presence of different bands with different algorithms. IET Signal Process 2013;7(8):720–30.
wavelength regions in the satellite images, efficiency of the [3] Iqbal MZ, Ghafoor A, Siddiqui AM. Satellite image resolution enhancement
using dual-tree complex wavelet transform and nonlocal means. IEEE Geosci
algorithms is effected and containing high resolution is one more Remote Sens Lett 2013;10(3):451–5.
cause of inefficiency as a result it leads to computational complex- [4] Kumar A, Bhandari AK, Padhy PK. Improved normalised difference vegetation
ity during enhancement. In case of satellite images, precisely index method based on discrete cosine transform and singular value decom-
position for satellite image processing. IET Signal Process 2012;6(7):617–25.
finding efficient enhanced result in the existence of inherent [5] Gupta V, Kumar L, Kumari U. Color satellite image segmentation using markov
uncertainty and ambiguity is a challenging task. At present one random field and multiresolutional wavelet transform. Int J Comput Appl Eng
is commonly faced with a situation that appeals appropriate Sci (IJCAES) 2011;1(1):75–9.
[6] Kim TK, Paik JK, Kang BS. Contrast enhancement system using spatially
enhancement. Enhancement leads to locate the area of interest, adaptive histogram equalization with temporal filtering. IEEE Trans Consum
detect, identify and assist in the satellite image segmentation. Electron 1998;44(1):82–7.
Such suitable and accurate multispectral remote sensing image [7] Chitwong S, Boonmee T, Cheevasuvit F. Enhancement of color image obtained
from the PCA–FCM technique using local area histogram equalization. Proc
enhancement can appreciably support the applications in numer-
SPIE 2002;4787:98–106.
ous fields ranging amongst agriculture, defense, geology, environ- [8] Gonzalez RC, Woods RE. Digital image processing. 2nd edition. Singapore:
mental science, etc. Usually, an image holds several ill-defined Pearson Prentice-Hall; 2002.
and ambiguous areas. The gray value allocated to a pixel is also [9] Kim, T, Yang, HS. A multidimensional histogram equalization by fitting an
isotropic Gaussian mixture to a uniform distribution. In: Proceedings of the
the typical reflectance of many kinds of land covers. Therefore, IEEE international conference image processing, vol. 8, no. 11, p. 2865–8; 2006.
appointing proper boosted features with firmness is an inherent [10] Weeks AR, Sartor LJ, Myler HR. Histogram specification of 24-bit color images
problem for satellite images. in the color difference (C–Y) color space. Proc. SPIE 1999;3646:319–29.
[12] Mukherjee J, Mitra SK. Enhancement of color images by scaling the DCT
coefficients. IEEE Trans Image Process 2008;17(10):1783–94.
[13] Mallat G. Theory for multi-resolution signal decomposition: the wavelet
Table 2 representation. IEEE Trans Pattern Anal Mach Intell 1989;2(7):674–94.
Thematic bands of NASA are as LANDSAT satellite [8]. [14] Donoho DL, Johnstone IM. Ideal spatial adaptation by wavelet shrinkage,.
Biometrika 1994;81(3):425–55.
Band Name Wavelength Characteristics and use [15] Donoho DL, Johnstone IM. Adapting to unknown smoothness via wavelet
shrinkage. J Am Stat Assoc 1995;90(432):1200–24.
no. (mm)
[16] Donoho DL. Denoising by soft thresholding. IEEE Trans Inf Theory 1995;41:
613–27.
1. Visible blue 0.45–0.52 Maximum water penetration
[17] Gao H, Bruce AG. WaveShrink with firm shrinkage. Stat Sin 1997;7:855–74.
2. Visible green 0.52–0.60 Good for measuring plant vigor [18] Gao H. Wavelet shrinkage denoising using the nonnegative garrote. J Comput
3. Visible blue 0.63–0.69 Vegetation discrimination Graph Stat 1998;7:469–88.
4. Near infrared 0.76–0.90 Biomass and shoreline [19] Fodder IK, Kamath C. Denoising through wavelet shrinkage: an empirical
mapping study. J Electron Imaging 2001.
5. Middle infrared 1.55–1.75 Moisture content of soil [20] Achim A, Bezerianos A, Tsakalides P. Novel Bayesian multiscale method for
6. Thermal 10.4–12.5 Soil moisture, thermal mapping speckle removal in medical ultrasound images. IEEE Trans Med Imaging
infrared 2001;20(8):772–83.
7. Middle infrared 2.08–2.35 Mineral mapping [21] Achim A, Kuruoghlu E. Image denoising using alpha-stable distributions in the
complex wavelet domain. IEEE Signal Process Lett 2005;12(1):17–20.

Please cite this article as: Bhandari AK, et al. Cuckoo search algorithm based satellite image contrast and brightness enhancement using
DWT–SVD. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2014.04.007i
 A.K. Bhandari et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 11

[22] Mohamad, M, Hamid, M. Ultrasound speckle suppression using heavy tailed [41] Bhandari, AK, Gadde, M, Kumar, A, Singh, GK. Comparative analysis of different
distribution in the dual tree complex wavelet domain. In: Proceedings of the wavelet filters for low contrast and brightness enhancement of multispectral
IEEE conference on wavelet diversity and design, p. 65–68; 2007. remote sensing images. In: Proceedings of the IEEE international conference
[23] Crouse MS, Nowak RD, Baraniuk RG. Wavelet-based statistical signal proces- on machine vision and image processing (MVIP), p. 81–6; 2012.
sing using hidden Markov models. IEEE Trans Signal Process 1998;46(4): [42] Bhandari AK, Kumar A, Padhy PK. Enhancement of low contrast satellite
886–902. images using discrete cosine transform and singular value decomposition.
[24] Mohamad, M, Hamid, M. Ultrasound speckle suppression using heavy tailed World Acad Sci Eng Technol 2011;79:35–41.
distribution in the dual tree complex wavelet domain. In: Proceedings of the [43] Kroeger, GC. Geologic remote sensing and multispectral image processing,
IEEE conference on wavelet diversity and design, p. 65–68; 2007. 1995.
[25] Gupta S, Chauhan RC, Saxena SC. Locally adaptive wavelet domain Bayesian [44] Mallat. SG. A theory for multiresolution signal decomposition: the wavelet
processor for denoising medical ultrasound images using Speckle modeling representation. IEEE Trans Pattern Anal Mach Intell 1989;11(7):674–93.
based on Rayleigh distribution. Proc IEEE Vis Image Signal Process 2005;152(1): [45] Ling Y, Ehlers M, Usery EL, Madden M. FFT-enhanced IHS transform method
129–35. for fusing high-resolution satellite images. ISPRS J Photogr Remote Sens
[26] Michailovich OV, Tannenbum A. Despeckling of ultrasound images. IEEE Trans 2007;61(6):381–92.
Ultrason, Ferroelectr Freq Control 2006;53(1):64–78. [46] Sverdlov A, Dexter S, Eskicioglu AM. Robust DCT-SVD domain image water-
[27] Bhuiyan, MIH., Ahmad, MO, Swamy, MNS. New spatial adaptive wavelet based marking for copyright protection: embedding data in all frequencies. In:
method for the despeckling of medical ultrasound image. In: Proceedings of Proceedings of the multimedia and security workshop, ACM Press; 2004. p.
the IEEE conference on circuits and system, p. 2347–2350; 2007. 66–174.
[28] Portilla J, Strela V, Wainwright MJ, Simoncelli EP. Image denoising using [47] Ayangar RV, Talbar NS. A novel DWT-SVD based watermarking scheme. In:
Gaussian scale mixtures in the wavelet domain. IEEE Trans Image Process
Proceedings of the IEEE International Conference on Multimedia Computing
2003;12:1338–51.
and Information Technology, 2010. p. 105–108.
[29] Pizurica A, Philips W. Estimating the probability of the presence of a signal of
[48] Demirel H, Ozcinar C, Anbarjafari G. Satellite Image contrast enhancement
interest in multiresolution single and multiband image denoising. IEEE Trans.
using discrete wavelet transform and singular value decomposition. IEEE
Image Process 2006;15(3):654–65.
Geosci Remote Sens Lett 2010;7(2):333–7.
[30] Rabbani H, Mansur V, Purang A, Saeed G. Speckle noise reduction of medical
[49] Demirel H, Anbarjafari G. Discrete wavelet transform-based satellite image
ultrasound images in complex wavelet domain using mixture priors. IEEE
resolution enhancement. IEEE Trans Geosci Remote Sens 2011;49(6):
Trans Biomed Eng 2008;55(9):2152–60.
1997–2004.
[31] Rabbani H. Image denoising in steerable pyramid domain based on a local
[50] Demirel H, Anbarjafari G. Image resolution enhancement by using discrete
Laplace prior. Elsevier J Pattern Recognit 2009;42(9):2181–93.
[32] Yu H, Zhao L, Wang H. Image denoising using trivariate shrinkagefilter in the and stationary wavelet decomposition. IEEE Trans Image Process 2011;20(5):
wavelet domain and joint bilateral filter in the spatial domain. IEEE Trans 1458–60 (May).
Image Process 2009;18(10):2364–9. [51] Demirel, H, Anbarjafari, G, Jahromi, MNS. Image equalization based on
[33] Zhang XP, Desai MD. Adaptive denoising based on SURE risk. IEEE Signal singular value decomposition. In: Proceedings of the IEEE international
Process Lett 1998;5(10):265–7. conference on computer and information sciences, vol. 1–5; 2008.
[34] Zhang, XP. State-scale adaptive noise reduction in images based on thresh- [52] Murty SP, Rajesh PK. A robust digital image watermarking scheme using
olding neural network. In: Proceedings of the IEEE conference on speech and hybrid DWT–DCT–SVD technique. Int J Comput Sci Netw Secur 2010;10(10):
signal processing, p. 1889–1892; 2001. 185–92.
[35] Zhang XP. Thresholding neural network for adaptive noise reduction. IEEE [53] Kakarala R, Ogunbona PO. Signal analysis using a multiresolution form of the
Trans Neural Netw 2001;12(3):567–84. singular value decomposition. IEEE Trans Image Process 2000;10(5):724–35.
[36] Nasri M, Pour HN. Image denoising in the wavelet domain using a new [54] Song L, Zhang S. Singular value decomposition-based reconstruction algorithm
adaptive thresholding function. Elsevier J Neurocomput 2009;72:1012–25. for seismic travel-time tomography. IEEE Trans Image Process 1999;8(8):
[37] Foi A, Katkovnik V, Egiazarian K. Pointwise shape-adaptive DCTfor high- 1152–4.
quality denoising and deblocking of grayscale and color images. IEEE Trans [55] Yang, XS, Deb, S. Cuckoo search via Lévy flights. In: Proceedings of the world
Image Process 2007;16(5):1395–411. congress on nature and biologically inspired computing (NaBIC), IEEE Pub-
[38] Yang J, Wang Y, Xu W, Dai Q. Image and video denoising using adaptive dual-tree lications, pp. 210–214; 2009.
discrete wavelet packets. IEEE Trans Circuits Syst Video Technol 2009;19(5): [56] Brown C, Liebovitch LS, Glendon R. Lévy flights in Dobe Ju/hoansi foraging
642–55. patterns. Hum Ecol 2007;35:129–38.
[39] Bhutada GG, Anand RS, Saxena SC. PSO-based learning of sub-band adaptive [57] Valian E, Mohanna S, Tavakoli S. Improved cuckoo search algorithm for
thresholding function for image denoising. Springer Signal, Image Video feedforward neural network training. Int J Artif Intell Appl (IJAIA) 2011;2(3).
Process (SIViP) 2012;6:1–7. [58] 〈http://www.gauss.ba/en/portfolio/satelitsko-snimanje/〉.
[40] Bhutada GG, Anand RS, Saxena SC. Image enhancement by wavelet-based [59] http://earthobservatory.nasa.gov/NaturalHazards/view.php?id=20046.
thresholding neural network with adaptive learning rate. IET Image Process [60] 〈http://www.satimagingcorp.com/gallery/geoeye-1-sahara-desert.html〉.
2011;5(7):573–82. [61] 〈http://earthobservatory.nasa.gov/IOTD/view.php?id=39125〉.

Please cite this article as: Bhandari AK, et al. Cuckoo search algorithm based satellite image contrast and brightness enhancement using
DWT–SVD. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2014.04.007i