Image Processing Reference
In-Depth Information
bands and sensitivity to unsensed phenomena. These properties are integral to human
vision. At present, human vision is far more sophisticated than we can hope to achieve with
a computer vision system. Infrared guided-missile vision systems can actually have difficulty
in distinguishing between a bird at 100 m and a plane at 10 km. Poor birds! (Lucky plane?)
Human vision can handle this with ease.
1.3.2
The neural system
Neural signals provided by the eye are essentially the transformed response of the wavelength
dependent receptors, the cones and the rods. One
model
is to combine these transformed
signals by addition, as illustrated in Figure
1.5
. The response is transformed by a logarithmic
function, mirroring the known response of the eye. This is then multiplied by a weighting
factor that controls the contribution of a particular sensor. This can be arranged to allow a
combination of responses from a particular region. The weighting factors can be chosen to
afford particular filtering properties. For example, in
lateral inhibition
, the weights for the
centre sensors are much greater than the weights for those at the extreme. This allows the
response of the centre sensors to dominate the combined response given by addition. If the
weights in one half are chosen to be negative, whilst those in the other half are positive,
then the output will show detection of contrast (change in brightness), given by the differencing
action of the weighting functions.
Logarithmic response
Weighting functions
Sensor inputs
log(p
1
)
w
1
log(p
1
)
p
1
log(p
2
)
w
2
×
log(p
2
)
p
2
Out
put
log(p
3
)
w
3
log(p
3
)
p
3
p
4
log(p
4
)
w
4
×
log(p
4
)
p
5
w
5
log(p
5
)
log(p
5
)
Figure 1.5
Neural processing
The signals from the cones can be combined in a manner that reflects
chrominance
(
colour
) and
luminance
(
brightness
). This can be achieved by subtraction of logarithmic
functions, which is then equivalent to taking the logarithm of their ratio. This allows
measures of chrominance to be obtained. In this manner, the signals derived from the
