Enhancement of Degraded Images by Natural Phenomena
Daily Daleno de O. Rodrigues¹, Anderson G. Fontoura¹, José R. Hughes Carvalho¹,
José P. de Queiroz Neto² and Renato P. Vieira¹
¹Instituto de Computação (IComp), Universidade Federal do Amazonas (UFAM), Manaus – Amazonas, Brazil
² Instituto Federal do Amazonas (IFAM), Manaus – Amazonas, Brazil
Keywords: Homomorfic Filter, Image Enhancement, Color Correction, Contrast Stretching, High-Pass Filtering, Bad
Condition Images, Information Lost.
Abstract: The efficiency of environmental monitoring through imagery data is strongly dependent on the quality of the
acquired information, despite weather conditions or other uncontrolled degradation factor. This article
describes a series of combined techniques of image enhancement to partially recover information “lost” due
to unfavorable operational conditions or natural phenomena, such as: fog, rainstorms, underwater dust (green
dust), poor illumination, etc. We based our approach on a process known as homomorphic filtering, which is
intrinsically related to the transformation from the spatial to the frequency domains, directly involving the
Fourier Transforms, followed by specific enhancement techniques, such as Clipping and Stretching.
Although, the use of these techniques separately, without the proper adaptation and coupling, can result in
damaging even more the image, the authors developed an efficient sequence of enhanced filtering able to
recover most of the affected information. Moreover, the proposed methodology proved to be generally
applicable to a large class of images in poor conditions, with a performance comparable to the methodology
used as benchmarks.
1 INTRODUCTION
The use of algorithms for image enhancement has
been a subject of study for decades. For the specific
application of environmental monitoring and
surveillance, the advent of software focused on image
manipulation for mitigating the effects of natural
phenomena drawn the attention from academic,
commercial and military research.
Therefore, the removal of the negative effects
promoted by phenomena like fog, rain, water, lack or
poor illumination, among others, becomes a
necessary requirement for a sort of applications.
Autonomous navigation of ground and aerial
systems, outdoor air or underwater remote sensing,
automatic object recognition, and active perception,
are just a few examples of situations where outdoor
imagery plays a critical role in a successful
application.
Studies on the area of image enhancement
indicate that in many working environments, several
factors influence negatively in scene visualization,
such as those present in the aerial environment, where
there are mist, fog, rain, smoke and hail (Oakley and
Satherley, 1998), (Liu et al., 2010) and (Tan et al.,
2007). These phenomena are mostly characterized by
the presence of aerosols and/or tiny water particles
suspended in air. Depending on its density they, in
one hand, may compromise considerably the original
characteristics of color and contrast of the images
and, on the other, hamper the ability of the observer
to perceive and interpret the information contained
therein.
The degradation of the visual quality of the
images fostered by these phenomena is modeled as a
function of density of particles in suspension along
the distance from the camera to the scene. The quality
index of visibility depends on the extension of the
scattering caused by these particles in the
phenomenon (Agarwal et al., 2013). Thereby, to
development a system to enhance the visibility of
acquired images in bad weather implies to mitigate
the effects caused by this scattering. Additionally, the
non-uniformity of scene illumination as captured by
the camera sensor negatively influences the image
quality. Due to the automatic camera adjustments, the
overexposing of one region, besides saturates that
region to white, also results in underexposing other
54
Daleno de O. Rodrigues D., G. Fontoura A., R. Hughes Carvalho J., P. de Queiroz Neto J. and P. Vieira R..
Enhancement of Degraded Images by Natural Phenomena.
DOI: 10.5220/0005266500540061
In Proceedings of the 10th International Conference on Computer Vision Theory and Applications (VISAPP-2015), pages 54-61
ISBN: 978-989-758-089-5
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
region, darkening part of the image. Therefore,
factors such as low natural illumination, lack of
equipment accessories (lens, flashes and filters), poor
capturing devices and inexperienced users deeply
affect the quality of the acquired image.
Researchers have devoted considerable efforts to
improve the acquired images independently to the
operational conditions of the data acquisition. In the
present work, the authors conduct a study of the
problem and propose a method to perform automatic
enhancement of the input (images or video) through
the use of filters with no prior knowledge of the
environment. In short, the contribution of this paper
is an effective combination of enhancement
techniques that besides recovering information and
improve contrast, cleanses the image from noise.
Comparison evaluation showed that the proposed
technique outperforms similar methods.
This work follows a very standard organization.
Section 2 shows a short description of correlated
works and contributions. On section 3, the authors
summarize the proposed method, and on section 4,
they describe the methodology and the evaluation
conditions of the experiments. Section 5 presents the
comparisons to others available in the literature and
discussions about the performance of the proposed
methodology. A psychovisual test, Mean Opinion
Score or MOS (Schettini et al., 2010), is performed
to measure the quality of correction. Finally, Section
6 concludes the paper.
2 RELATED WORKS
The image enhancement methods in the literature can
be classified into two main categories, based on the
physical model of the image formation and based on
the enhancement of the image using subjective and
objective criteria to produce a visually pleasing image
(Gonzalez and Woods, 2010).
The work of Zhai et al. (2007) provided an
overlapped modification of histogram equalization on
sub-block to improve images affected by the
phenomenon of fog.
Panetta et al. (2008) proposed an algorithm for
image enhancement with variable illumination to
enhance the local contrast, and also to keep the details
of the edges. These authors also proposed an
algorithm of multi-histogram equalization in the HSV
color space to segment the image, allowing a rapid
and efficient correction of non-uniform illumination.
The work of (Iqbal et al., 2007) developed a
model that improves the techniques and methods of
perception of underwater images based on slide
stretching applied both in the RGB as the HSV color
spaces.
Kalia (Kalia et al., 2011), investigated different
pre-image processing techniques that can affect or
improve the performance of the SURF (Speed-Up
Robust Features) detector, and proposed a new
method named IACE (Image Adaptive Contrast
Enhancement). They modified the technique of
contrast enhancement, adjusting it according to the
statistics of image intensity levels. Equation (1)
enables the estimation of the change in intensity
levels P
out
according to the intensities of the levels P
in
of the image to be enhanced.
,
,
.
(1)
Where a is the lower possible intensity level of the
image. b is its corresponding counterpart. c is the
lowest level of the threshold intensity of the original
image for which the number of pixels in the image is
less than 4% and, d is the intensity level of the upper
threshold for which the cumulative number of pixels
is greater than 96%. These thresholds are used to
eliminate the effect of outliers, improving the
intrinsic details of image while maintaining the
contrast ratio. However, the values of P
out
should be
in the range [0, 255].
The results of this algorithm are very promising.
The relative performance of IACE method is better
than the method proposed by (Iqbal et al., 2007), in
terms of time needed to process improvement and
complete matching image. The contributions of
studies (Kalia et al., 2011) and (Iqbal et al., 2007)
occurs in the sense of making use of contrast
stretching algorithms both in RGB and HSV, besides
the results indicate prosperity by applying thresholds
at the end of the stretching intensities.
At last, (Schettini et al., 2010), proposed a method
for image enhancement based on a location-
dependent exponential image correction. The
technique aims to correct images that have both
underexposed and overexposed regions
simultaneously. To avoid artifacts, the bilateral filter
is used as a mask in the exponential correction.
Depending on the characteristics of the image (driven
by histogram analysis), an automatic step of tuning
parameters is introduced, followed by stretching,
trimming, treatment and preservation of color
saturation. His contribution is on the use of cropping
and stretching, as well as the correction of color
saturation stage.
EnhancementofDegradedImagesbyNaturalPhenomena
55
3 PROPOSED METHOD
3.1 Basic Block-Diagram
The authors proposed a three-step improvement
method, depicted in the block-diagram of Figure 1.
The method is applied to the input image converted
into HSV color space, due to the fact that the human
eye maximizes the perception of color of objects in
the scene.
The first step corrects the non-uniformity of the
illumination, making it as more homogeneous as
possible throughout the scene. This step works on the
information present in both underexposed and
overexposed regions. The second step involves
contrast enhancement, intended to enhance image
details. Finally, the third step minimizes color
saturation changes between the input and output
images, to vivid them as close as possible to the actual
colors. The following sections explain in details each
step.
Figure 1: Block diagram of the proposed method.
3.2 Illumination Correction, Fast
Fourier Transform and High Pass
Filtering
As explained by (Padmavathi et al., 2010), the
formation of images can modeled as the product
matrix generated by the intensity of illumination and
reflectance of objects in the scene, that is:
,
,
∙
,
(2)
where , represents the captured image, ,
the illumination and , the reflectance of the
objects in the scene. One needs to consider that
illumination tends to vary slower (low frequency),
throughout the image when compared to the
reflectance, characterized by abrupt changes,
especially in the edges (high frequency). Therefore,
by suppressing lower frequency components, while
reinforcing medium and high frequencies, one would
address the issue. However, the necessary Fourier
transform is a non-linear operation, and any attempt
of filtering will include both illumination and
reflectance matrices. According to eq. (3):
,
,
,
(3)
According to (Delac, 2006), (Gonzalez and Woods,
2010) and (Padmavathi et al., 2010), there are five
steps to obtain the corrected image.
Step 1: Apply the logarithm operator to linearize the
process on grayscale (HSV if color) image (eq. (4).
,
log
,
log
,
log
,
(4)
Step 2: Apply Fast Fourier Transform, as in eq. (5).
,
log
,
log
,
→
,
,
,
(5)
Step 3: Filtering by a Butterworth high-pass filter. If
, is processed with a high-pass filter ,,
we obtain eq. (6).
,
,
∙
,
,
∙
,
,
∙
,
(6)
Where , is the result of image on frequency
domain with the high-pass filter.
Step 4: Apply the Inverse Fourier Transform to go
back to the spatial domain (eq. (7)).
,
,
,
′,
(7)
Assuming:
,
,
,
(8)
In addition:
,
,
,
(9)
Step 5: Apply the exponent
ial operator in all image to revert the effects of
logarithm on step 1. Since , is constructed as
the log ,, the inverse of , leads to the
desired result on eq. (10):
,
,
→
,
,
∴
,
,
(10)
The , represents the homomorphic filtered
image of
,
. If the intensities are high, a second
or third additional enhancement technique can be
applied (Toth, 2011).
In this work, we used the Cooley-Tukey (Diniz et
al., 2014) implementation of the Fast Fourier
Transform at Step 2. The overall code complexity of
this step is equal to
(Weeks, 2012).
VISAPP2015-InternationalConferenceonComputerVisionTheoryandApplications
56
Butterworth high-pass filtering was used to obtain a
sharper image, by attenuating low frequency
components without affecting the high-frequency
information in the Fourier. The eq. (12) shows the
filter expression:
,
1
1
,
⁄
(11)
Where n is the degree of transform, constant
is the
threshold of the filter, and , is the distance
from point , to the center of the frequency
spectrum, given by:
,
2
⁄
2
⁄
(12)
Where is the size of input image in frequency
domain (in that case, , of eq. (5)).
3.3 Contrast Enhanced (CLAHE,
Clipping and Stretching)
This step involves contrast improvement, and is
intended to retrieve image details emerged from a
deeper analysis of the histogram intensities before
and after application of homomorphic filtering. In this
scenario, we realized that even after a better
occupation of gray levels, improving overall contrast
is not satisfactory, since low-quality images have
compression artifacts and noise in darker areas are
maximized. These negative characteristics were also
perceived in (Schettini et al., 2010) and (Kalia et al.,
2011).
The contrast enhancement is divided into sub-
steps. First, it transforms the model RGB to HSV and
then applies the technique of Contrast Limited
Adaptive Histogram Equalization (CLAHE) in
channel V. Its use aims to facilitate the recovery of
details in dark areas of pictures, as well as prevent the
loss of information due to excessive gloss, because its
histogram is not limited to a particular region.
We noticed that CLAHE promote the
maximization of noises. To avoid this, we also
applied a contrast threshold to eliminate the effect of
extreme values and improve the intrinsic image
details, while maintaining the contrast ratio.
However, the output pixel values are redistributed in
the range [0, 255], also reported by (Sakaue et al.,
1995, Kalia et al., 2011). This practice significantly
improve the picture quality, but, increases the
complexity of the method to 2 or.
To determine the contrast threshold that will be
used to limit the intensities, we will consider the
saturation levels at the ends of the histogram, more
specifically, those distributed outside the interval of
97%. This percentage is experimentally obtained, and
proved to be the best threshold indicator after several
trials.
At last, the contrast enhancement step uses the
technique of contrast stretching on both S and V
channels in HSV color space. After that, we used the
same technique on RGB color space to stretch the
dynamic range of the image intensities. The intension
is to provide brighter colors and make dark colors
even darker. This procedure provides a gain in
quality, i.e., to stretch the histogram so that the full
dynamic range be better distributed along the same
range. This dynamic range is the range between the
minimum and maximum intensity value obtained
after applying the threshold.
3.4 Color Saturation
As already mentioned, images that suffer from
degradation due to environmental phenomena
generally have turbid or opaque colors. To retrieve
them and finalize the processing chain, we applied the
same formula suggested by Sakaue (Sakaue et al.,
1995) and used in (Schettini et al., 2010). This idea
seeks to minimize the color variation between the
input image and the output. The transformations are
applied to each RGB channel model producing new
values R’, G’ e B’ obtained as:
1
2
1
2
1
2
(13)
where V’ is the value of the intensity of illumination
obtained after the illumination correction and contrast
enhancement with histogram clipping, as discussed
earlier. The values corresponding to V, R, G e B are
obtained in the input image.
4 EXPERIMENTS
4.1 Evaluation Methodology
The experiments were implemented in C++ and
OpenCV library set, and simulated on a two-core @
2.5GHz personal computer with 8GB RAM. The
proposed method was applied in twelve color images
that show degradations resulting from various natural
anomalies that Nature can adversely promote in the
atmosphere or underwater. The images have different
sight distances and different rates of turbidity. They
EnhancementofDegradedImagesbyNaturalPhenomena
57
are classified into different scenarios, including
scenes that present problem of non-uniformity of
illumination, problems in the atmosphere, and
problems in underwater environments. For such
classification, four images were selected and
submitted to each of the three methods in evaluation.
Details of the images that are beyond the limits of
visibility are not considered as regions to be
reclaimed. The proposed method is subjected to the
color space RGB and HSV in various channels, as
was shown in Figure 1.
The assessment of image quality is performed
subjectively from the point of view of the observer.
In each trial of the experiments, a pair of images is
available for viewing each of the 30 volunteer
evaluators. This pair of images consists of two
versions of the same scene and an evaluator was
randomly invited to respond subjectively indicating
which image was his favorite. Each version of a scene
was compared with all other versions of the same
scene, representing 24 pairs (12 images with 2
combinations each). Overall, 720 trials were
conducted and the results are explained in Chapter 5.
4.2 Evaluation Conditions
To perform the psychovisual evaluation, the images
were judged to be shown in an interface based
desktop applications. We adopted an LED monitor of
a 14’’ notebook with resolution of 1366 x 768 pixels
corresponding to 111.94 dpi. The refresh rate is 60
Hz. The lighting is typical of office and the ambient
light levels were kept constant between numerous
sessions. The distance between the observer and the
monitor was about 60cm. All original scenes used
were subsampled and scaled to fit a square 600 x 600
pixels.
5 RESULTS
We submitted the images to the proposed
methodology and, then, compared the results
obtained with others techniques. Figure 2, shows the
result when running the steps of illumination
correction and contrast enhancement.
We noticed that in (a), due to low light in the
scene, many information were "hidden" from a
typical human eye. Through the homomorphic filter,
many details appear in (b), however, one ca note a
slight saturation in overexposed region of the
background image. In (c), the result of (b) was
combined with techniques CLAHE, Clipping and
stretching to make the light become better distributed
Figure 2: In (a) original image under low light; In (b) image
with simple homomorphic filter and (c) image with the
proposed correction method to non-standard lighting,
consisting of homomorphic filtering, CLAHE, Clipping
and Stretching.
in the scene and the edges more prominent. Despite
the image (c) apparently be better, the complexity of
the operation is still high (in
), but it is a major
advances compared with
. A future work will
be to reduce this complexity to
.
Table 1 presents the results obtained of three
different scenarios in pairs of images, original (left
side) and after the application of the proposed method
(right side). From a subjective evaluation, we find
significant improvements in all scenarios, whether in
environments with serious illumination issues or in
environments with issues arising from acts of nature,
and both in atmosphere and in underwater
environments, with low and high turbidity.
Table 1: Comparison between before and after the proposed
method applied.
Fig. Original Proposed Method
Images with poor illumination
1
2
3
A
B
C
VISAPP2015-InternationalConferenceonComputerVisionTheoryandApplications
58
Table 1: Comparison between before and after the proposed
method applied (Cont.).
Fig. Original Proposed Method
Images with poor illumination
4
Foggy images
5
6
7
8
Underwater images
9
10
Table 1: Comparison between before and after the proposed
method applied (Cont.).
Fig. Original Proposed Method
Underwater images
11
12
In order to compare with others techniques, we first
conducted a psychovisual experiment in which
viewers were asked to choose the better perceived
picture of a pair. Based on the preferences of viewers,
we obtained an average score of opinions. The
content preference score is calculated for each
method, including the original scene (Identity). The
highest score goes to 100. First place got 51.25 points,
corresponding to the proposed method. Regarding the
other two methods, the technique of (Schettini et al.,
2010) is on second place with 25.41 points and the
model of (Iqbal et al., 2007) with 22.5 point, on third
position. Table 2 shows the results.
Table 2: Average Opinion Score (MOS) experiment.
Method
(Result Image)
Preference
Score
%
Identity 6 0.834
Proposed 369 51.251
Schettini et al. 183 25.417
Iqbal et al. 162 22.523
One may realized that each method has a better
performance in specific aspects. When considering
the correction factor of uniformity of illumination, the
algorithm of (Iqbal et al., 2007) is the one with the
lowest score, however, it gets good recovery in
underwater imagery. The algorithm of (Schettini et
al., 2010), in short, have as good results as the our
proposed method, however, it does not promote good
outcomes for underwater images, being the
determining factor of its second place rank, as
presented in Table 2. This scenario determines that
applying algorithms in contrast saturation and
brightness of channels HSV color space, as well as in
each of RGB channels, promotes not only the
EnhancementofDegradedImagesbyNaturalPhenomena
59
improvement of the contrast, but also makes color
recovering feasible.
Measuring the quality enhancement applied to an
image after enhancement process is often very
difficult to accomplish. So far applies to subjective
evaluation criteria of improvement in the image. In
literature some objective metrics that aim at
estimating the brightness and contrast of the image,
such as Entropy (H), Absolute Mean Brightness Error
(AMBE) and Enhancement Measure (EME). The use
of the values of these metrics need certain care, since
it does not necessarily correlate improved quality in
terms of contrast enhancement (Schettini et al., 2010).
Analyzing the values of these metrics for image 8
of Table 1, the values of H indicate better occupation
of all intensity levels in the histogram. This is
associated with a visually pleasing image, positive
considering the balanced conditions of image
acquisition. The values of this metric for that image
are 6.9384, 7.6334, 6.8803 and 7.3561. Representing
the result of the original image, the proposed method,
the method of (Schettini et al., 2010) and of (Iqbal et
al. 2007), respectively. It is noteworthy that in this
case, the highest value of Entropy in fact represented
a nice image. However, it does not always happen in
other cases. Analogous behavior is found in the study
of (Schettini et al., 2010).
The AMBE is the average distance from the
original brightness, i.e., the difference in the average
intensity level of the gray scale, and new original
image. In a procedure improvement, if not always
aims to preserve the original brightness of the scene,
given the problems of uniformity of illumination.
(Schettini et al., 2010) states that preserve the original
brightness does not always mean preserve the natural
appearance of the image. Additionally, the original
images are strongly underexposed and/or
overexposed, we expect a high value AMBE,
indicating that the quality could be improved. For
example, the steps resulting from the applied
illumination correction and contrast enhancement of
the image 1, shown in Table 1, obtains a correction
value AMBE equals to 49.52. On the other hand, in a
correct exposure images or pictures obtained in dark
scenes (overnight), it is expected that our method
does not significantly alter the average brightness.
The metric values for image 2 of Table 1 with this
characteristic are, the proposed method equals 19.39,
AMBE (Schettini et al, 2010) equal to 20.46 and
AMBE (Iqbal et al, 2007) equal to 34.38.
The EME approximates an average contrast in the
image by dividing the image into no overlapping
blocks, defining a measure based on minimum and
maximum intensity values in each block and
averaging them. By this metric, high values should
indicate regions with high local contrast, while values
close to zero, should correspond to homogeneous
regions. If improvement method introduces noise in
such homogeneous regions, a higher value of EME
will be obtained, and possibly not correspond to an
improvement in image quality. As an example, the
values of this metric in image 10 of Table 1: Proposed
Method = 9.2414, Schettini method = 1.7403 and
Iqbal method = 5.4655. In this scenario values, on the
one hand, we can see that our method was the one
with the highest value; this is due mainly to the use of
CLAHE. Moreover, this value is not necessarily as a
negative factor which might compromise the quality
of the circumstantially improvement obtained.
6 CONCLUSIONS
This article presents a method of enhancement of
images degraded by natural phenomena which allows
the improvement of visibility from a single input
image without using any information of their training
model. The goal is to improve the visual quality of
distant objects on the scene. From the results, one can
notice that most of the intensity of degrading
phenomena are minimized, providing a better contrast
enhancement, brightness and color brightening
compared to other techniques. The authors found the
methodology promising for showing good results
with low computational cost and useful for their
application in various systems working in outdoor or
submerged environments.
When we compared the proposed solution with
other well-known method in the literature, we find a
correct increased dynamic range in both regions of
low and high brightness of an image, preventing the
common loss of quality due to artifacts, desaturation,
low luminance and grayish appearance. The Mean
Opinion Score (MOS) was used to evaluate the
performance of different contrast correction methods
of color images. The proposed method reached the
highest scores.
The method adequately performed in all three
different scenarios, especially when compared to
others. Its performance with images of underwater
environments, where the method achieved the higher
points, was especially interesting. However, in future
works, we intend to improve the algorithm optimizing
the parameter estimation values from the model of
image formation, such as attenuation and diffusion
coefficients that characterize the turbidity of scene
and depth of a given object in image.
VISAPP2015-InternationalConferenceonComputerVisionTheoryandApplications
60
ACKNOWLEDGEMENTS
This work was sponsored by FAPEAM through
Support Program of Postgraduate HR Training of
Interior for the State of Amazonas (FAPEAM No
1270/2012), Project ARTES (FAPEAM No.
114/2014), CNPq/SETEC-MEC Call 015/2014, and
PROMOBILE/Samsung under the terms of Brazilian
federal law No. 8.248/91 (Unisol No 93.00.00).
REFERENCES
Agarwal, V., Khandelwal, S., Goyal, D., Sharma, J., Tiwari,
A., 2013. Two-Pass Adaptive Histogram Based Method
for Restoration of Foggy Images. The LNM Institute of
Information Technology. Jaipur, India.
Delac, K., Grgic, M., Kos, T. 2006. Sub-Image
Homomorphic Filtering Technique for Improving
Facial Identification under Difficult Illumination
Conditions. University of Zagreb, Faculty of Electrical
Engineering and Computing. Budapest, Hungary. 4p.
Diniz, Paulo Sergio R. et al. 2014. Processamento Digital
de Sinais, Bookman. Porto Alegre, Brazil, 2
nd
edition.
Gonzalez, Rafael C., Woods, Richard C., 2010.
Processamento Digital de Imagens. Pearson. São
Paulo, Brazil, 3
th
edition. p. 22-64.
Iqbal, K., Salam, R. A., Osman, A., Talib, A. Z., 2007.
Underwater Image Enhancement Using an Integrated
Colour Model. International Journal of Computer
Science. Penang, Malaysia.
Kalia, Robin, Lee, Keun-Dong, R. Samir B. V., Sung-Je,
Kwan, Oh, Weon-Geun, 2011. An analysis of the
effect of different image preprocessing techniques on
the performance of SURF: Speeded Up Robust
Feature. 17
th
Frontiers of Computer Vision (FCV),
Korea-Japan Joint Workshop. Daejeon, South Korea.
Liu, Huiyan, He, Wenzhang, Liu, Rui, 2010. An Improved
Fog-degrading Image Enhancement Algorithm Based
on the Fuzzy Contrast. International Conference on
Computational Intelligence and Security. Beijing,
China.
Oakley, John P., Satherley, Brenda L., 1998. Improving
Image Quality in Poor Visibility Conditions Using a
Physical Model for Contrast Degradation. IEEE
Transactions on Image Processing, Vol. 7, No. 2.
United Kingdom, London.
Padmavathi, Ganapathi, et al., 2010. Comparison of Filters
used for Underwater Image Pre-Processing.
Department of Computer Science, Avinashilingam
University for Women, Coimbatore, TN, India. 8p.
Panetta, Karen A., Wharton, Eric J., Agaian, Sos S., 2008.
Human visual system-based image enhancement and
logarithmic contrast measure, IEEE Transactions on
Systems, Man and Cybernetics - Part B: Cybernetics, p.
174–188.
Parker, J. R., 2011. Algorithms for Image Processing and
Computer Vision, Wiley. 2
nd
edition. p. 277-280.
Sakaue, S., Tamura, A., Nakayama, M., Maruno, S., 1995.
Adaptive gamma processing of the video cameras for
the expansion of the dynamic range, IEEE Trans.
Consum. Electron, p. 555–562.
Schettini R., Gasparini, F., Corchs, S., Marini, F., Capra,
A., Castorina, A., 2010. Contrast image correction
method. Milano, Italy.
Tan, Robby T., Pettersson, Niklas, Pettersson, Lars, 2007.
Visibility Enhancement for Roads with Foggy or Hazy
Scenes. IEEE Intelligent Vehicles Symposium.
Istanbul, Turkey.
Toth, Daniel, Aach, Til, Metzler, Volker, 2011.
Illumination–Invariant Change Detection. Institute for
Signal Processing, University of Lubeck. Germany. 5p.
Weeks, Michael, 2012. Processamento Digital de Sinais
Utilizando Matlab e Wavelets, LTC. Rio de Janeiro,
Brazil, 2
nd
edition.
Zhai, Yi-Shu, Xiao-Ming, Liu, 2007. An improved fog-
degraded image enhancement algorithm, International
Conference on Wavelet Analysis and Pattern
Recognition. Beijing, China, p. 522–526.
EnhancementofDegradedImagesbyNaturalPhenomena
61
