Researchers from the Optics Department at the
U. of Granada have developed a new algorithm for the capture of high dynamic
range (HDR) images which reduces the time of capture or the level of noise in
the resulting image. Beyond the field of photography, this new development can
also be applied to artificial vision systems, medical imaging, control quality
systems in assembly lines, satellite images, vigilance systems and assisted or
automatic driving systems for vehicles, etc.
This
research has been published in the journal Applied
Optics, and it has facilitated the generation of an algorithm which, based
on the way in which each camera responds to light, adapts the times of capture
of different images in an automatic and instantaneous fashion, in order to
reduce the total capture time, thus adapting to the conditions and needs of
each captured scene.
Although
the most striking and visually attractive application of HDR techniques is in
the field of photography, these techniques are particularly relevant in robotic
vision systems. "In the same way in which the human visual system has a
high dynamic range,
artificial vision systems also need this sort of techniques to behave in a similar,
or even more sophisticated, way vis à vis our human visual system",
according to the author of this research, Miguel ÁngelMartínez Domingo, from
the Optics Department, U. of Granada.
The
exposure time is the time in which the camera sensor is exposed to light during
image capture. In a short exposure, the dark zones of the scene appear
overexposed (black) in the final image. In a long exposure, by contrast, the
most luminous zones in the scene appear saturated or burnt out (black)
Sub exposed or saturated zones
"In
general, in any scene captured nowadays, and even though our camera
automatically adjusts the time of exposure, there will always be sub exposed or
saturated zones. This happens due to the fact that the range of luminosities
(or dynamic range) which the sensor in any conventional camera can capture
correctly in a single shot is smaller than the actual dynamic range of the
scene itself. This is where HDR capture techniques make sense", according
to Martínez Domingo.
What
really matters is not just that the final image turn out to be nice or
realistic for a human eye: it can be important, for instance, to highlight with
a low degree of noise the details of a very bright component and another dark
one in an integrated circuit. Consequently, the levels of noise in the
resulting image are incompatible with the total time of capture for the
different images required.
This
algorithm developed by U. of Granada scientists allows for the optimization of
the balance between a reduced time of capture and a lower level of noise in the
resulting image. The level of noise becomes lower the more exposures are
captured to make up the HDR image, but the use of many different exposures
would excessively increase the time of capture. "The new algorithm does
not just adapt itself to the camera we are using and the scene we are
capturing, but also to the specific needs of the application we are
developing," according to Martínez Domingo.
Consequently,
"for the first time an HDR image capture algorithm adapts itself to any camera,
any scene, and any application, in an easy and optimal way, without having to
use complex optic systems or non-conventional sensors architectures".
