Computer Vision News - August 2017

28 / 56

Derya Akkaynak is a postdoctoral
researcher at the University of
Haifa, jointly with the
Interuniversity Institute for
Marine Sciences in Eilat.
Derya (whose name means 'ocean' in
both Persian and Turkish) told us about
her work which she is presenting
today, titled “What Is the Space of
Attenuation Coefficients in
Underwater Computer Vision?”.
The poster she is presenting is about
improving color reconstruction and
color acquisition in images that are
collected underwater with underwater
robots or divers. Derya, who is an
oceanographer and mechanical
engineer (not a computer scientist by
training), is working on understanding
how light propagates underwater and
how it gets captured on camera
sensors, to find out how we can
compensate for the colors that are
lost, in an accurate and objective way.
Her paper leverages decades worth of
data from optical oceanography to
improve underwater computer vision
algorithms, bridging the two fields.
A main challenge of her work is to
validate the mathematical models they
build, and see if they actually work in
an underwater setting. This requires
many dives, a lot of equipment, and a
lot of hardware and sometimes things
can go wrong, or the results are
unexpected, and then they have to go
back to the model and adjust it. “So it’s
a constant iteration between work on
the computer, and work in the sea ”,
Derya says, which is different to most
Computer Vision fields, where work is
mostly done on a computer.
Talking about previous work, Derya
explains to us that there is an existing
system of equations for underwater
image formation, and that everybody
uses these equations. However, Derya
and her co-authors looked at how
these equations were derived and they
found that due to two simplifying
assumptions there are errors that are
introduced that affect people’s work in
color reconstruction. So instead of
using what was commonly accepted,
Derya and her co-authors questioned it
and were then able to highlight the
weaknesses, and offer a better
solution.
We asked Derya if she thinks that we
can one day see underwater images
just like we see things in normal life.
She hypothesizes that this might be
possible with a lot of specialised
equipment, because we now
understand very well what happens to
light and how cameras capture light
underwater.
“The other deep
in computer vision”
Derya Akkaynak
28
What Is the Space of Attenuation Coefficients in Underwater Computer Vision?