IRJET-Design and Implementation of Haze Removal System

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 05 | May 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2411
Design and Implementation of Haze Removal System
Deepa. S. N1, Aishwarya. H2, Anuradha. M. G3
1, 2 Student, Dept. of Electronics and Communication Engineering, JSSATEB, Karnataka, India
3Assistant Professor, Dept. of Electronics and Communication Engineering, JSSATEB, Karnataka, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Haze is one of the natural phenomena where
turbid medium conceals scenes diminishes color contrast and
reduces visibility. Removal of haze from these images is an
important challenge, where the dehazed images find
applications in automobiles for better visibility and in
photography. The main aim lies in improvising the human
visibility in the bad weather conditions. In the paper, a dark
channel algorithm and median filtering is used to remove the
haze from an image. It is seen that the proposed algorithm
works for all types of haze in daytime.
Key Words: Dehazing, Airlight, Scene Radiance
1. INTRODUCTION
Low visibility condition is a frequent occurrence. Poor
visibility leads to flight delays, diversion, car accidents and
also pollution has increased the problem and providing it to
be fatal. The task is to deal with situations which reduce
human performance due tolowvisibility. Thehumaneyehas
a resolution of some 324 to 576 pixels depending upon the
angle of vision, but in the cases of haze or other weather
conditions visibility is impaired. The main ambiguity
involved is the atmospheric light which degrades the image,
by scattering the light reflected fromthesceneofpointwhen
a hazy image is captured. The dehazingprocessisperformed
by using the dark channel prior algorithms. The visibility
matrix if an images is obtainedwhichfoundtobeimprovised
than previous methods. Multiple image based approachwas
used for defogging byNarasimhanandNayar[1].Polarization
based vision through haze approach used polarization
properties of an image for dehazing by Sehechner[2].The
Novel based fast defogging method from a single image of a
scene based on a fast bilateral filtering method, but this
involved of complexity of obtaining linear functions of a
number of input image pixels[3].Image depth based
methods, demand some depth information of the images
from the user inputs[4].
In this paper we propose a dark channel algorithm
for dehazing an image. Dark channel is for outdoor images
which is completely a statistical approach. So, the dark
channel is estimated for hazy images in which few pixels
possess low intensity with respect to any one color channel
i.e., RGB plane. The dark pixel estimation provides rough
information about the thickness of haze and also about the
amount of atmospheric light or air light responsible fordark
pixel formation. Further from the results of dark channel,
transmission map is estimated which helps to obtain or
recover scene radiance. Noise removal involves usage of
median filters and visibility metric for the haze free image is
calculated by necessary equations.
2. DARK CHANNEL ALGORITHM
The approach involves mainly the application of dark
channel algorithm to obtain haze free image with high
visibility and color contrast. This also involves the
application of Hybrid medianfilter[5]toremovetheimpulse
noises present in the haze free image, which adds on to the
higher visibility metric. The hazy images are described by
eqn (2.1),
Fig 1: Physical structure of a hazy imaging model.
I(x) =J(x) t(x) +A (1-t(x)) (2.1)
Where, I is the intensity of hazy image, J is the scene radians
of haze free image, A is atmospheric light or air light and t is
the transmission map. The aim is to obtain J of a hazy image.
The term J(x) and t(x) is the multiplicative deformity of the
scene radians and A (1-t(x)) is the additive deformity, air
light means the amount of light that is scattered when
reflected from the scene point.
When the atmosphere is uniform, the transmission map t
is given as,
(2.2)

Where β is the scattering coefficient of the atmosphereandd
is the distance from the camera (capturing point) to the
scene of point.
Fig 2: Block diagram of the proposed method
Fig 2 shows the block diagram. The method proposed
to recover the haze free image is as follows:
2.1. Read the hazy image
To consider a image of a particular scene of point
which contains dense haze, whose color contrastandquality
is been degraded. The dark channel algorithm is applied for
the same, in order to obtain a haze free image.
2.2. To calculate dark channel prior
The dark channel algorithm is considered for
outdoor haze free images. In most of the non-sky regionsthe
intensity of the pixels is very low or zero in atleast one color
channel.
Conventionally, the dark channel of a hazy randomimage Jis
given by Jdark ,
(2.3)
Where Jc is a color channel of J and Ω(x) is the size of local
patch centered at pixel x. The dark channel of a hazy image
obtained is by eqn (2.3) the result of two minimum
operators. These two minimum operators are independent
of each other.
By observation, it can be inferred that most of the pixels of
outdoor image possess very low intensity value or zero in J.
(2.4)
The low intensity in these dark channels is mainly due to
three factors…, 1. Shadows of various objects like trees,rock
etc.…2. Colorful objects like flowers, water bodies etc.…3.
Black objects (object black by nature). The dark channel of a
hazy image possesses higher intensity in the regions of
dense haze, and lower intensity in the regions of lesser
dense. This observation provides rough estimation about
thickness of haze.
2.3. Selection of Haze Opaque Region
In order to estimate the atmospheric light, this
causes the degradation of image with respect to color
contrast and visibility. We first determine the Haze opaque
regions, the regions which have nearly less or no haze, these
regions provide the actual light reflected from sceneofpoint
because of no haze being present for attenuation or
scattering, this involves human interaction for selecting the
same. The top 0.1% of the brightest pixels are considered as
haze opaque regions.
2.4. Atmospheric light Estimation
Haze attenuates the reflected lightfromthesceneof
point and some amount ofatmosphericlightisblended. Most
haze opaque regions are considered as atmospheric light A.
But the proposed method provides a much efficient way to
estimate atmospheric light A with respect to each channel
which accordingly increases the accuracy given by,
(2.5)
(2.6)
(2.7)
Where, AR is Atmospheric light of red color channel,
AG Atmospheric light of green color channel.
AB Atmospheric light of blue color channel.
Fn = filter (Jc, I0)
Fd = filter ((Jc-Fn)2, I0)
2.5. Transmission Map Estimation
Transmission map provides the amount of
deformity in an image due to atmospheric light. If t(x) is the

amount of light which is not scattered and reaches the
camera when a hazy image is captured. By normalizing the
hazy eqn (2.1) as:
(2.8)
Where c represents color channel. The dark channel is
calculated on both sides of eqn (2.8) by using minimum
operators:
(2.9)
By dark channel prior algorithm, wehavetheapproximation
of eqn (2.3) using the same in eqn (2.9) the expression for
transmission map is given as,
(2.10)
In dark channel approximationsthe coloroftheskyisalmost
close to the color of the atmosphere, so t(x) tends to zero,
which means that true transmission proves that the method
could efficiently deal with both sky and non sky regions. In
normal days the atmosphere possessessomeamountofhaze
and is not haze free. So, when we completely remove haze
from an image it loses its naturality, in order to preserve the
same some amount of haze is retained in the final image.
(2.11)
Value lies in between 0 and 1. The value of is
application based we fix it as 0.07 in the paper.
Patch size Ω(x) is one of the important parameters in the
algorithm, it is estimated that larger the patch size the dark
channel obtained is much better and is less precise for small
patches. But by choosing larger patch sizes the final image
obtained would be too saturated, so the patch size we
consider here for estimation is 15*15.
2.6. Recover Scene Radiance.
By the estimation of atmospheric light and the transmission
map and substituting in eqn (2.1) we could obtain scene
radiance, when true transmission occur thedirectdeformity
term tends to zero. The final scene radiance is given as,
(2.12)
The scene radiance obtained directly may possess noise so
the lower bound of eqn is restricted to t0.
2.7. Hybrid Median Filter
After the application of the scene radiance equation
the haze free image is more prone to impulse noise. The
noise occurred is due to transmission errors, storage
problems etc. So, to eliminate this noise Median filters are
used. Hybrid median filter is one of them which eliminates
noise in much better way by considering neighboring pixels
in both cross and plus format. Firstly, the pixels in cross
neighborhood are considered and median of them is
obtained and then the median of the pixels present in plus
neighborhood are obtained. Now the filter compares two
median values with the centre pixel and obtains the final
median value.
Fig 3. Cross and plus format of HFM
3. EXPERIMENTAL RESULTS
In this section a hazy image of size 1200*1200 is considered
in fig (4), by the application of dark channel algorithm. The
dark channel of the image in fig (5)providerough estimation
of haze. fig (6) helps in selection of haze opaque region for
atmospheric light estimation. Fig(7)isthetransmission map
which explains the amount of light that reached camera
without scattering. Fig (8) gives the haze free image and the
complete haze free image after noise removal by hybrid
median filter is given in fig (9).
Fig 4. Hazy arbitrary image

Fig 5. Dark Channel Image
Fig 6. Selection of Haze Opaque Region
Fig 7. Transmission map
Fig 8. Haze free image
Fig 8. HFM Image
By performing the necessary calculations, wetrytoestimate
the visibility metric.
4. CONCULSION
This paper demonstrates the importance of dehazing an
image. By the combination of Dark Channel Algorithm and
Hazy Imaging Model, we obtain much sophisticated and
efficient approach. The equation for atmospheric light
provides much more realistic value for A and the visibility
metric is found to be high due to this approach. Now for the
image of size 1200*1200 the visibility metric obtained is
25.073.
REFERENCES
[1] S. G Narasimhan and S. K Nayar, “Vision and the
Atmospheric” Intc’l Computer Vision, vol.48, pp. 233-
254, 2002.
[2] Y. Y Schechner, “Instant dehazing of images using
polarization,” Proc IEEE Conf. Computer Vision and
Pattern Recognition, vol. 2, pp.1984-1991,2006.
[3] R. Fattal, “Single Image Dehazing,” Proc. ACM
SIGGRAPH ’08, 2008.
[4] R. Tan, “Visibility in Bad Weather from a Single Image,”
Proc. IEEE Conf. Computer Vision and Pattern
Recognition, June 2008.
[5] “A Hybrid Median Filter For EnhancingDimSmall Points
Targets and Its Fast Implementation” by Qingyun Yang
in 2011.
[6] J. Kopf, B. Neubert, B. Chen, M. Cohen, D. Cohen-Or, O.
Deussen, M. Uyttendaele,andD.Lischinski,“DeepPhoto:
Model-Based Photograph Enhancement and Viewing,”
ACM Trans. Graphics, vol. 27, no. 5, pp. 116:1-116:10,
2008.
[7] E.B. Goldstein, Sensation and Perception. Cengage
Learning 1980.
[8] A.J. Preetham, P. Shirley, and B. Smits, “A Practical
Analytic Model for Daylight,” Proc. ACM SIGGRAPH ’99,
1999.
[9] C. Tomasi and R. Manduchi, “Bilateral Filtering for Gray
and Color Images,” Proc. Sixth IEEE Int’l Conf. Computer
Vision, p. 839, 1998.

IRJET-Design and Implementation of Haze Removal System

More Related Content

What's hot (20)

Similar to IRJET-Design and Implementation of Haze Removal System (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET-Design and Implementation of Haze Removal System