Concluding Remarks - Vision with Direction - page 365

machine vision, e.g., the generalized Hough transform, with directional processing.
In the text, the belief that most vision problems can be solved by systems mimick-
ing the known processing of human vision in their frontend is not concealed. In this
euphoria, it would nevertheless be a serious mistake to pretend that all methods and
principles to achieve human visual intelligence are covered, not least because there
are too many of these yet to be discovered.
The dimension explosion in the visual pathways is not an unknown phenomenon
in the mathematical treatment of signals. It goes under the general notion of feature
extraction . It is nearly routinely demanded in all advanced decision support systems
because each dimension brings a simplification, not available in the original signal.
In comparison, it is worth noting that an image filtered through a particular direc-
tional filter is more predictable than the original, and therefore less complex. The
filtered image, in its essence 1D, e.g., see [163], is more likely constant in the tune-
on direction than the orthogonal direction because of the specific filtering. Several
chapters of the topic treat therefore principles that are also found in statistics liter-
ature. However, the visual signals have unique properties which can allow a vision
system to limit its feature extraction to directional processing as a resource manage-
ment strategy to cope with multiple environments.
In English and in numerous other languages, the notion of vision is also synony-
mous with qualified imagination about the future. However, predicting the future of
visual signal analysis as a science is a feat even larger than solving the problems of
vision. Accordingly, what I expressed in this topic cannot be anything but my vision
of the matter, which may not be perceived as neutral—it is necessarily directional.