Image Processing Reference
In-Depth Information
Taken together, the central vision is well equipped to analyze sharp details be-
cause its cells in the cortex have receptive fields that are capable to quantify high
spatial frequencies isotropically, i.e., in all directions. This capability is gradually
replaced with spatial low-frequency sensitivity at peripherial vision where the cell
receptive fields are larger. In a parallel fashion, in the central vision we have cells
that are more suited to analyze slow moving patterns, whereas in the peripherial vi-
sion the fast moving patterns can be analyzed most efficiently. Combined, the central
vision has most of its resources to analyze high spatial frequencies moving slowly,
whereas the peripheral vision devotes its resources to analyze low spatial frequen-
cies moving fast. This is because any static image pattern is equivalent to sinusoidal
gratings, from a mathematical viewpoint, since it can be synthesized by means of
these.
7
The spatial directional selectivity mechanism is a result of interaction of cells in
the visual pathway, presumably as a combination of the LGN outputs which, from
the signal processing point of view, are equivalent to time-delayed outputs of the
retinal ganglion cells. The exact mechanism of this wiring is still not well understood,
although the scheme suggested by Hubel and Wiesel, see [113], is a simple scheme
that can explain the simple cell recordings. It consists in an additive combination of
the LGN outputs that have overlapping receptive fields. In Fig. 1.8, this is illustrated
for a bar-type simple cell, which is synthesized by pooling outputs of LGN cells
having receptive fields along a line.
A detailed organization of the cells is not yet available, but it is fairly conclusive
that depthwise, i.e., a penetration perpendicular to the visual cortex, the cells are or-
ganized to prefer the same spatial direction, the same range of spatial frequencies,
and the same receptive field. Such a group of cells is called an
orientation column
in
the neuroscience of vision. As one moves along the surface of the cortex, there is
lo-
cally
a very regular change of the spatial direction preference in one direction and oc-
ular dominance (left or right eye) in the other (orthogonal to the first). However, this
orthogonality does not hold for long cortical distances. Accordingly, to account for
the spatial direction and ocular dominance changes as one moves along the surface,
a rectangular organization of the orientation columns in alternating stripes of ocular
dominance is not observed along the surface of the cortex. Instead, a structure of
stripes, reminiscent of the ridges and valleys of fingerprints, is observed. Across the
stripes, ocular dominance and along the stripes, spatial direction preference changes
occur [222].
The direction, whether it represents the spatial direction or the motion, is an
important feature for the visual system because it can define the boundaries of objects
as well as encode texture properties and corners.
8
Also, not only patterns of static
images but also motion patterns are important visual attributes of a scene because
object background segregation is tremendously simplified by motion information by
motion information compared to attempting to resolve this in static images. Likewise,
7
We will introduce this idea in further detail in Chap. 5.
8
The concept of direction can be used to represent texture and corners in addition to edges.
In Chaps. 10 and 14 a detailed account of this is given
