Single Underwater Image Enhancement

with a new optical Model

by H Wen - 2013 ISCAS

Outline
1.
2.
3.
4.

Underwater Image formation optical model

New Underwater Dark channel prior
Estimate the transmission map
Recover the scene radiance from the
observed image

Haze-Image Optical Model

Haze-image formation model
= + ( ())
direct transmission

I(x): observed image

J(x): scene radiance
A: the global atmospheric light
t(x): transmission map

airlight

Theoretical basis
Light dissemination:

water absorbs the longer wavelength of red and scatters the

blue and violet when visible light disseminates in it.

New image optical model

Haze-image model:
= + ( ())
Underwater-image model:

c: denotes a color channel which can be red, green, blue.

: transmission rate
: scattering rate
In water, there exists absorption, transmission and scattering.

Fig.1 Frame diagram of the method

Dark channel prior

The traditional dark channel prior:
The new underwater dark channel prior:

Background light estimation

The background of an underwater scene tends to be
blue or green.

p=(i, j) is the pixel location where we get the background light from
the input image. As shown in the following picture (red circles).

Scattering rate estimation

To estimate :
Assume the transmission and scattering rate in a local patch is
constant. ( *, + )

Transmission estimation
To estimate :
Since the attenuation of both blue and green light results from
scattering:
Since the farther the object from the camera, the more energy
of red light is absorbed, the intensity of the red color channel
of an image can qualitatively reflect the transmission rate of
red channel.

Results