Enhanced Vision of Hazy Images Using Improved Depth Estimation and Color Analysis

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 03 Issue: 02 | Feb-2016 www.irjet.net p-ISSN: 2395-0072
© 2016, IRJET | Impact Factor value: 4.45 | ISO 9001:2008 Certified Journal | Page 579
ENHANCED VISION OF HAZY IMAGES USING IMPROVED DEPTH
ESTIMATION AND COLOR ANALYSIS
Mr. Prasath P 1,Mr. Raja G 2
1Student, Dept. of comp.sci., Dhanalakshmi Srinivasan Engineering College,Tamilnadu,India.
2Assistant Professor, Dept. of comp.sci., Dhanalakshmi Srinivasan Engineering College, Tamilnadu,India.
Abstract
Images are captured during inclement weather
conditions such as fog, sand, and mist, that images are
called as hazy images. Those images are frequently feature
degraded visibility and unwanted color cast effects. Due to
these effects, original image may not be clear. In such
situation Laplacian-based visibility restoration
approaches usually cannot adequately restore images due
to poor estimation of haze thickness and the persistence of
color cast problems. In proposed system, Enhanced refined
transmission technique is used to solve effectively
inadequate haze thickness estimation and alleviate color
cast problems. It improves the performance quality of
systems such as object recognition systems, obstacle
detection systems, video surveillance systems, intelligent
transportation Systems. Experimental results via
qualitative and quantitative evaluations demonstrate that
the proposed method can dramatically improve images
captured during inclement weather conditions and
produce results superior to those of other state-of-the-art
methods.
Key Words: Laplacian-based visibility restoration,
Enhanced refined transmission.
1. INTRODUCTION
Diverse weather situations such as haze, fog,
smoke, rain, or snow will cause multifaceted visual
effects of spatial or temporal domains in images or
videos [1–3]. Such artefacts may appreciably
humiliate the performances of outdoor vision
systems relying on image/video feature extraction
[4] or visual attention modelling [5–7], such as
event detection, object detection, tracking, and
recognition, scene analysis and classification, image
indexing and retrieval [8].Images or video bear
from lack of quality taken under such conditions,
unless the hazy appearance is needed for artistic
reasons. Visibility restoration [9] refers to different
methods that seek to reduce or remove the
degradation that have occurred while the digital
image was being obtained. The reasons of the
degradation can be factors like blurring due to
camera miss focus, relative object-camera motion,
relative atmospheric turbulence and various others.

Fig .1 a) Input Haze Image b) Output Haze Free Image
In this paper, we discuss the degradations duo to
bad weather like fog, haze, rain and snow in an
image. The key reason of degradation of picture
quality of outside screen in the mist and fog climate
condition is for the most part the diffusing of a light
before arriving at the camera because of these
extensive amounts of suspended particles (e.g.
haze, dimness, smoke, impurities) in the weather.
This influences the typical working of automatic
monitoring system, outdoor recognition system and
intelligent transportation system. By the use of haze
removal methods of picture we can improve the
stability and strength of the visual framework. Haze
is a case of the opaque medium (e.g., particles and
water droplets) in the atmosphere, which will
humiliate outdoor images due to atmospheric
absorption and scattering [8].Its removal is an
excessive undertaking as fog relies on upon the
unknown scene depth data. Fog effect is the
function of distance between camera and object.
The haze removal approach might be divided into
two classifications: image enhancement and image
restoration. Image enhancement classification
excludes the reasons of haze humiliating picture
quality. This method loses a portion of the data in
regards to image additionally enhance the contrast
of fog picture. Image restoration firstly studies the
physical process of image imaging in foggy climate
[9].Several algorithms have been anticipated to
boost the quality of images taken under foggy
environment, focusing for instance on visibility.
2. BACKGROUND
There are several methods for estimating the haze
contribution in a single image. One of the most
successful is the enhanced refined transmission,
proposed by He et al.,4 and is used as the basis for
haze estimation in this paper. Here we briefly
summarize the approach.
The enhanced refined transmission is derived
assuming a noise-free image with the following
image formation model:
I(x) = R(x) t(x) + a∞ (1 − t(x)) (1)
where I(x) represents the hazy image captured by
the digital camera; J(x) represents the scene
radiance, which can be regarded as a haze-free
image; A represents the global atmospheric light; x

represents each pixel location of the incoming
image; and t(x) represents transmission map, which
is used to describe the non scattered light between
the observed objects and the digital camera. Note
that the first term J(x) t(x) and the second term A(1
− t(x)) of (1) represent the direct attenuation and
the air light, respectively [26]. The decay of scene
radiance J(x) can be described by direct
attenuation, which is subject to medium and scene
depth information.
Moreover, the air light value usually suffers from
scattering and absorption by atmospheric particles,
resulting in scene colour variation. When the
atmosphere considered here is assumed to be
homogenous, the transmission map t(x) can be
expressed as
t(x) = e−βd(x) (2)
Where β represents the scattering coefficient of the
atmosphere, and d(x) represents the scene depth
information between the observed object and the
digital camera’s obtained by simply subtracting the
dark channel of the normalized image from 1. The
scaling parameter, w, takes a value from 0 to 1, and
corresponds to the amount of haze left in the image.
After the initial haze estimate is obtained, a
refinement step is required to suppress halo
artefacts. He et al.4use the Matting Laplacian.12
Although this is not the quickest solution, it
provides visually satisfactory results, and so is the
process used in this paper. Finally, the dehazed
image is usually recovered by simple inversion of
Eq. (1), solving for R.
3. SYSTEM MODEL
In this section we consider the existing
system design and the proposed system.
3.1 Existing System
In Existing the dark channel prior is to estimate
scene depth in a single image and it is estimated
through get at least one color channel with very low
intensity value regard to the patches of an image.
The transmission map will be estimated through
atmospheric light estimation. The key to the dark
channel prior is the observation that natural haze-
free outdoor images are generally well textured,
and contain a variety of colourful objects. As a
consequence, most patches will contain one or
more pixels with very low intensity in at least one
of the colour channels. These dark pixels can be
attributed to dark objects, shadows, or objects that
are primarily a combination of only one or two of
the RGB colour channels. With this observation in
mind, one can construct the so called “dark
channel” of an image, which can be expressed
mathematically as a minimum value operation in
patches around the target pixel:
(3)
where Idark(x) represents the”dark channel” of
image I at pixel location x, Ic is a colour channel of
image I, and y ∈ Ω(x) signifies all pixels y in a local
patch around x. If applied to a haze-free image, the
above observation yields:
(4)
In contrast, hazy images contain an additive
atmospheric light component, yielding:
(5)

Since t(x) is essentially constant in a local window
and a∞ is usually close to white (meaning it won’t
significantly affect the dark channel value), the dark
channel prior effectively identifies the relative haze
content throughout an image. This can aid in
estimating the atmospheric light, a∞. Following Ref.
4, we estimate this component as the brightest RGB
intensities in the hazy image among the pixels
corresponding to the top 0.1% brightest dark
channel locations, i.e. the regions where haze is
most dominant.
3.2 Proposed System
In our method the haze removal was done using
visibility restoration of single hazy images using
color analysis and depth estimation with enhanced
refined transmission technique. The hazy removal
technique divided into three categories such
additional information approaches, multiple image
approaches, single-image approaches. The first two
methods are expense one and high computational
complexity. Recently single image approach is used
for this dehazing process because of its flexibility
and low cost. The restoration model is proposed
with utilization of median filter and adaptive
gamma correction technique. This approach
overcomes the problems such as color distortion,
artifacts and insufficient depth information.
3.2.1 Module description
3.2.1.1 Median filtering:
In noise removal process, initially we convert the
image in gray. And then we filter the noise from the
image. In Filtering we are applying Median filtering
to our input image. Gaussian filtering is often used
to remove the noise from the image. Here we used
wiener2 function to our input image. Median filter
is windowed filter of linear class by its nature is
weighted mean named after famous scientist Carl
Gauss because weights in the filter calculated
according to Gaussian distribution.
3.2.1.2 Adaptive Gamma Correction
Adaptive Gama correction is used to enhance
contrast in digital image that improve the
brightness of dimmed image via the gamma
correction of luminance pixels. Automatic
modifying histogram and enhancing contrast and
Improve brightness of dimmed images, Gama
correction often simply gamma, is name of a
nonlinear operation used to code and decode
luminance values in video or still image system.
Gama correction is, in the simplest case, defined by
the following power law expression.
Vout=ANin
ɤ
3.2.1.3 Depth Map Estimation
Depth estimation refers to the set of techniques and
algorithms aiming to obtain a representation of the
spatial structure of a scene and to obtain a measure
of the distance of each point of the seen scene.
Depth estimation in computer vision and robotics is
most commonly done via stereo vision, in which
images from two cameras are used to triangulate
and estimate distances. However, there are also
numerous monocular visual cues. Such as texture
variations and gradients, defocus, color/haze.
3.2.1.4 Visibility Restoration
Visibility restorations refer to different method that
aim to reduce or remove the degradation that have
occurred while the digital image was being
obtained. Visibility Restoration module can
effectively recover visibility in the restored image
and high-quality image can be generated.

4. EXPERIMENTAL RESULTS
This section explores the qualitative and
quantitative evaluations to comprehensively
compare the proposed method to the other ERT-
based methods, including the methods of He et al.
[2], Xie et al. [7], Xu et al. [28], and Jin et al. [12].
The evaluations are achieved by using
representative image databases, including Flickr
[16], Picasa [18], and Google Images [20], for a total
sample set of 1586 images. In addition, all
compared approaches were implemented by using
the C programming language on an Intel Core i3
3.07-GHz processor with 2 GB of RAM, running a
Windows 7 operating system. Quantifying the
restored images is a very difficult task. This is
because a real-world haze-free reference image has
not been validated for quantification of restored
images. In general, the objective metrics used for
quantitatively estimating the efficacy of the
restoration results can be divided into two
categories, i.e., reference methods and non-
reference method. In our experiments, we perform
the local min operator using Marcel van Herk’s fast
algorithm [8] whose complexity is linear to image
size. The patch size is set to 15 × 15 for a 600 × 400
image. In the soft matting, we use Preconditioned
Conjugate Gradient (PCG) algorithm as our solver. It
takes about 10-20 seconds to process a 600×400
pixel image on a PC with a 3.0 GHz Intel Pentium 4
Processor. Figure e show our haze removal results
and the recovered depth maps. The depth maps are
computed using Equation (2) and are up to an
unknown scaling parameter β. The atmospheric
lights in these images are automatically estimated
using the colour analysis method. As can be seen,
our approach can unveil the details and recover
vivid colour information even in very dense haze
regions.
(a) (b) (c)
(d) (e)
Fig.2 a) Input Hazy Image b) Median filter c) Gamma correction d) Depth map e) Output Haze free image

CONCLUSION
It is the unique method for the Outdoor
Object recognition systems. It is done using Color
Analysis and Depth Estimation with Enhanced
refined transmission. It is a simple and efficient
method. The System has the better performance
and less computation speed than the existing
system. Future enhancement of this Project can be
implemented to Advanced Driver Assistance
System or Obstacle Detection systems by also
adding sand and mist removal features.
REFERENCES
[1] Fang, Faming, Fang Li, Xiaomei Yang, Chaomin
Shen and Guixu Zhang(2014), "Single image
dehazing and denoising with variational
method", IEEE International Conference on
Image Analysis and Signal Processing (IASP),
pp. 219-222.
[2] Huang S.C, F. C. Cheng, and Y. S. Chiu(2015),
“Efficient contrast enhancement using adaptive
gamma correction with weighting distribution”
IEEE Trans. Image Process., vol. 22, no. 3, pp.
1032–1041.
[3] He, Kaiming, Jian Sun and Xiaoou Tang(2014),
"Single image haze removal using dark channel
prior.",IEEE Transactions on Pattern Analysis
and Machine Intelligence, vol.33, no. 12, pp.
2341-2353.
[4] Long, Jiao, Zhenwei Shi and Wei Tang(2014),
"Fast haze removal for a single remote sensing
image usingdark channel prior", International
Conference on Computer Vision in Remote
Sensing (CVRS), pp.132-135.
[5] Tan, Robby T(2014), "Visibility in bad weather
from a single image” IEEE Conference on
Computer Vision and Pattern Recognition,
CVPR, pp. 1-8.
[6] Tarel, J-P. and Nicolas Hautiere(2013), "Fast
visibility restoration from a single color or gray
level image",12th International Conference on
Computer Vision, pp. 2201-2208.
[7] Ullah, E., R. Nawaz and J. Iqbal(2013), "Single
image haze removal using improved dark
channel prior",Proceedings of International
Conference on Modelling, Identification &
Control (ICMIC), pp. 245-248.
[8] Xu, Haoran, Jianming Guo, Qing Liu and Lingli
Ye(2012), "Fast image dehazing using improved
darkchannel prior", IEEE International
Conference on Information Science and
Technology (ICIST), pp.663-667.
[9] Huang and B. H. Chen, “Highly accurate moving
object detection in variable-bit-rate video-
based traffic monitoring systems,” IEEE Trans.
Neural Network. Learn. Syst., vol. 24, no. 12, pp.
1920–1931, Dec. 2013.
BIOGRAPHY
Mr.P.Prasath received the B.E-CSE
degree from Sri Krishna College of
Engineering and Technology,
Coimbatore. He is currently doing
his M.E-CSE degree in
Dhanalakshmi Srinivasan
Engineering College, Perambalur,
Tamilnadu, India.

Enhanced Vision of Hazy Images Using Improved Depth Estimation and Color Analysis

More Related Content

What's hot (20)

Similar to Enhanced Vision of Hazy Images Using Improved Depth Estimation and Color Analysis (20)

More from IRJET Journal (20)

Recently uploaded (20)

Enhanced Vision of Hazy Images Using Improved Depth Estimation and Color Analysis