presentation1 (2)
TRANSCRIPT
Presented by
Pooja G Bidwai
Sources of underwater image distortion are
1. Light Scattering
2. Color Change
Light scattering lowers the visibility and contrast of
captured image.
Color change leads to the varying degrees of
attenuation.
Fig.1 Hazing and bluish effect caused by light scattering and color change
in underwater images.
To enhance the underwater image Wavelength
compensation and dehazing algorithm is most
appropriate(WCID).
Enhanced visibility and superior color fidelity can be
obtained by WCID algorithm.
Image dehazing helps to restore clarity of underwater
images.
Haze is caused by suspended particles such as sand,
minerals in lakes, river and oceans.
Capturing image underwater is challenging due to haze
caused by light.
Using hazy image formation model, image
formed at camera can be given as:
X= point in underwater scene,
= image captured by camera
= scene radiance at point x.
= residual energy ratio.
= homogeneous background light
Residual energy ration can be given as:
Normalized residual energy ratio depends on light
wavelength transmitted & it can be shown as:
Different wavelengths of light are attenuated at diff rates in water
Incident light traverses from surface of water reaching
the image scene covering range from D through D+R,
where R corresponds to image depth range.
Amount of residual light w(x) formed after wavelength
attenuation can be:
(3)
Light J(x) emanated from point x is equal to the amount
of illuminating ambient light Reflected i.e.
Is the reflectivity of point x
(4)
Eq. 1 can be modified as :
(5)
This equation incorporates light scattering during
propagation from object to camera d(x), and
wavelength attenuation along both D(x) and d(x).
Haze increases with distance; so, haze can be useful to
determine d(x).
Using single image haze removal and dark channel
prior d(x) can be determined-
If point x belongs to a part of foreground then
Jdark(x) ->0
But background light is usually assumed to be pixel
intensity with highest brightness,
Thus foreground and background intensities are known
and further it is used to segment background and
foreground.
Depth map of fig.1obtained by estimating d(x), i.e. distance between object
and the camera using dark channel prior
Artificial lights are often supplemented to avoid
insufficient lighting in underwater photographic
environment.
Artificial light source can be detected by comparing
luminance difference of foreground and the
background.
Based on depth map derived:
◦ Area = foreground if d(x)> σ
◦ Area = background if d(x)<= σ
Influence of artificial lighting is a function of amount
of luminance by light source and surface reflectance of
objects.
Illuminated by artificial light source, intensity of foreground appears
Brighter than that of background.
After detecting the artificial light source, can be
removed by subtraction from main image formation
equation:
After removal of artificial light in fig 1. is shown in right panel
Of split screen, whereas original one is in left panel.
Haze can be removed by subtracting term (1-tλ(x)).Bλ
in eq.1:
Image obtained after eliminating haze, light scattering and color change .
Image obtained after processing with (a)WCID (b) dark-channel based Dehazing (c) chromatism-based dehazing (d) histogram qualization
RESULT
WCID results in superior haze removal and color
balancing capabilities over dehazing and histogram
equalization.
Highest SNR values are obtained.
Performance of WCID is most robust through different
water depths.
John Y.Chinag and Ying-Ching Chen, “Underwater image enhancement by
wavelength compensation and dehazing,”IEEE transaction on image processing,
vol.21,no.4,April 2012.
K. He, J. Sun, and X. Tang, “Single image haze removal using DarkChannel Prior,”
in Proc. IEEE CVPR, vol. 33, no.12,Dec.2011.
K. Lebart, C. Smith, E. Trucco, and D. M. Lane, “Automatic indexing of
underwater survey video: algorithm and benchmarking method,” IEEE J. Ocean.
Eng., vol. 28, no. 4, pp. 673–686, Oct. 2003.
Y. Y. Schechner and N. Karpel, “Recovery of underwater visibility and structure by
polarization analysis,” IEEE J. Ocean. Eng., vol. 30, no. 3, pp. 570–587, Jul. 2005.