Contrast-Based Stereoscopic Images Dehazing
Yimin Qiu
1
1
School of Information Science and Engineering
Wuhan University of Science and Technology
Wuhan, China
qiuyimin@wust.edu.cn
Shiqian Wu
2,3
2
School of Machinery and Automation
Wuhan University of Science and Technology
3
Hubei Collaborative Innovation Center for Advanced
Steels
Wuhan, China
shiqian.wu@wust.edu.cn
Abstract—As human eyes perceive scenes with slightly
different angles, fog effect is referred to the function of the
distance between camera and objects. In this paper, a novel
contrast-based dehazing algorithm is proposed by using
stereoscopic images. The proposed algorithm first decomposes
the disparity map in a given fog-and-haze stereo pair with digital
wavelet transformation (DWT). Then, the contrast sensitivity
function (CSF) is employed to adjust the image contrast. To
measure the contrast, the cost function, which consists of the
DWT CSF mask and wavelet contrast measurement is proposed.
Results on a variety of real hazy images demonstrate that the
proposed approach significantly improves hazy image quality.
Especially, the proposed method has fast speed so that it can be
implemented in real-time applications.
Keywords—stereoscopic images; dehazing; contrast
enhancement; wavelet transformation;HVS
I. INTRODUCTION
Fog and haze are inseparable and unavoidable phenomena
which usually degrades images of outdoor scenes. The turbid
medium, such as particles or water droplets reduces
atmospheric visibility. The fog-and-haze case appears in many
metropolises of China recently, not only giving people serious
harm, but also leading to traffic accidents. Hence, it’s
necessary to improve the fog-and-haze images by using image
processing technologies [1].
Visually, the effects of fog-and-haze lead to a significant
contrast loss in the degraded images. In the literature, a few
methods have been proposed to restore such degraded images
by relying on image enhancement approaches. The
commonly-used enhancement methods include gray-
transformations by means of nonlinear functions, histogram-
based techniques in spatial domain [1-3]
and methods
operating in the frequency domain, such as the wavelet
transformation [4]
. Recently, some novel single image
dehazing algorithms have been developed [5-11]. Tarel et al [5]
proposed a method based on linear operations which has a fast
visibility restoration, but it requires many parameters for
adjustment. Fattal [6] proposed an independent component
analysis method. This method estimates hazy scenes using
color information and accordingly it is not appropriate in
dense fog images because of fog’s colorless. Tan [7] removed
fog by maximizing local contrast of the image, but the restored
images is liable to over-saturation. Kopf et al [8] proposed a
method based on three-dimensional model, which is
application-dependent and expert interaction is required. He et
al [9] proposed a method based on dark channel prior and soft
matting. But this algorithm is invalid while the scene objects
are as bright as the atmospheric light. For the methods
working on single image dehazing, the enhanced images
always have annoying artifacts, such as blurred edges, halo
artifact or distortion. Consequently, methods using multiple
images have been proposed [10-14]. In [10,11], polarization
filtering is employed to dehaze using two or more images
which have the same scene but different degrees of
polarization. One drawback of this method is the limitation in
image acquisition, i.e., it cannot be applied to dynamic scenes
[5]. More constraints in capturing multiple images were
applied in [12-14].
Different from the aforementioned conventional multi-
image dehazing, in which the same scene but in time-lapse
images are captured, we capture two images of the same scene
simultaneously but in different views, i.e., using stereoscopic
system. Therefore, image acquisition is very easy and the
proposed method can applied to either static scenes or
dynamic scenes. Since human eyes perceive objects with
slightly different angles, it is reasonable to use stereoscopic
images to distinguish objects by three-dimensional vision [15],
and help outdoor visual cognition capabilities, especially
under fog-and-haze environment [1]. In this paper, fog effect
is referred to the function of the distance between camera and
objects by estimating depth map of the scenes. A novel
contrast-based dehazing algorithm is proposed by using
stereoscopic images. Wavelet transformation is used to
increase the contrast of the image prior to fog removal.
Contrast Sensitivity Function (CSF) is then employed to
determine the contrast of images through CSF-mask weighting
and increases contrast of the fog removed image. The masking
function of CSF is adaptive according to the image content.
Experimental results show that the proposed algorithm
achieves better results than other existing algorithms.
This paper is organized as follows. In Section II,
characteristics of human visual system used in the proposed
algorithm is discussed. In Section III, the proposed dehazing
algorithm is described in detail. Section IV demonstrate
experimental results, and followed by conclusion in Section V.
597
978-1-4799-8389-6/15/$31.00
c
2015 IEEE