Image Processing Reference
In-Depth Information
cies), whereas those in the central area, i.e., basically M- and L-cones, respond best
to high spatial-variations. The fineness (spatial frequency) at which a photoreceptor
has its peak sensitivity decreases with increased eccentricity of the receptors. At the
periphery, where we find rods and S-cones, the photoreceptors respond to low spa-
tial variations (silhouettes) whereas the central vision dominated by M- and L-cones
responds better to high spatial variations.
2.3 Neuronal Operations and Color
Color perception is the result of comparisons, not direct sensor measurements. The
amount of photons with a narrow range of wavelengths reflected from a physical sur-
face changes greatly as a function of the time of the day, the viewing angle, the age of
the viewer,..., etc., and yet humans have developed a code that they attach to surfaces,
color . Human color encoding is formidable because, despite severe illumination vari-
ations (including photon wavelength composition), it is capable of responding with
constant color sensation for the viewed surface. This is known as color constancy .
It has been demonstrated by Land's experiments [145] that the color of a viewed
patch is the result of a comparison between the dominant wavelength of the reflected
photons from the patch and those coming from its surrounding surface patches.
The signals coming from the L-, M-, and S-cones of the retina, represented by
L , M , and S here, arrive at the two lateral geniculate nucleus (LGN) areas of the
brain. At the LGN the signals stemming from the 3 cone types in the same retinal
proximity are presumably added and subtracted from each other as follows:
L +
M
Lightness sensation
L
M
Red-green sensation
L + M
S
Blue-yellow sensation
The ˜
represents a local, weighted spatial summation of the respective cone type
responses [113]. The local window weighting is qualitatively comparable to a 2D
probability distribution (summing to 1), e.g., a “bell”-like function ( Gaussian ). The
positive terms in the three expressions have weight distributions that are much larger
at the center than those of the negative terms. Accordingly, the net effect of L
·
M is
a center-surround antagonism between red and green, where red excites the center
as long as there is no green in the surround. If there is green in the surround the
response attenuates increasingly. This signal processing functionality is found among
parvocellular cells in layers 4-6 of the LGN , called ( r + g
)-cells. However, the
L
M above can result in negative values if
˜ L< M .In
mathematical expression
that case another group of cells, the ( g + r
)-cells which are also found among
the parvocellular cells, will deliver the negative part of the signal
L
M , while
the ( r + g
)-cells will be inactive. The ( g + r
)-cells function in the same way
as ( r + g
)-cells except that they are excited by green in the center and inhibited
by r in the surround. Accordingly, ( r + g
)- and ( g
r +)-cells together implement
L
S results in an antagonism between blue and yellow.
This scheme is presumably implemented by two groups of parvocellular LGN cells,
M . Likewise, L +
M
Search WWH ::




Custom Search