|
Introduction
Background
Algorithms
Test Images
Testing Interface
Image Comparison
Methods
Results
Conclusion
Possible
Extensions
References
Appendix I
Appendix II |
Background
Light
The light that comes into our eyes is a combination of the reflection
and refraction properties of the objects we are looking at and the spectral
power distribution (SPD) of the light that illuminates the object of
interest. This light has been scattered on an object in two ways: it has
either been reflected in the surface in what is called interface reflection,
or it has entered the object and bounced back and forth between the
particles until it eventually exited the object at some angle, which is
called body refraction. Taking all this into account, the SPD coming from an
object c(l)
into our eyes can be described as:
where b(l)
is the body reflectance, i(l)
is the interface reflectance and e(l)
is the SPD of the illuminant. The interface reflection is almost uniform for
all incident wavelengths, meaning that the light reflected in an object’s
surface has the same wavelength distribution as the illuminant. Hence, i(l)
is a constant – let’s call it i – and the formula becomes:
 |
Human color vision
The SPD of the light from an object as shown above is clearly a function of
the illuminant, meaning that the light coming from an object under one
illuminant will not have the same wavelength distribution as the light
coming from the same object under a different illuminant. Still, we say that
an orange is orange, no matter if we are looking at the object in an indoor
environment under incandescent light or outside in bright daylight.
The explanation to why we always perceive orange as orange is that our
visual pathways have an amazing ability to correct for the illuminant. Two
different SPD’s cause different amounts of isomerizations in our three types
of cones and thereby give rise to different L, M and S color signals. These
signals are interpreted as being the same thanks to a process called
approximate color consistency. Our brain gets information about the
illuminant from all the light that comes into our eyes and uses this
information as a decoding scheme for the color signals, a scheme that
compensates for the illuminant.
Digital images
The sensor of a camera, or any other type of digital image acquisition
device, is reached by essentially the same light as an eye in the same
position. The photons reaching the sensor will induce a response which
carries information about the energy (or wavelength) and the number of
photons. Hence, the sensor’s response to light is similar to the human
cones’. However, as stated above, this color information is a combination of
the illuminant and the color of the objects the light is coming from, and
the sensor itself – as our eyes – has no way of correcting for the
illuminant. Therefore, the raw data from camera sensors often give images a
color cast or include colors that do not look natural. To try to correct for
this misrepresentation, we apply post-processing algorithms that use the
information available in the image to remove the color cast and make the
colors appear natural to a human eye. Some of these algorithms find an
estimate of the illuminant and then apply a color transformation that
approximately divides it out (eg white balancing), others work in a more
indirect manner by making assumptions about the color distribution that the
image should have (eg grayworld, scale by max). This process of trying to
correct for different illuminants and various camera color distortions is
called color balancing.
|
