Image Processing Reference
In-Depth Information
18
Concluding Remarks
The term
vision
encompasses much more than registering the distribution of light by
our eyes or by a camera. It refers to a process that analyzes what has been captured,
so as to facilitate actions and inferences. In this topic, we anatomized the observed
light distribution by using multidimensional signal analysis tools and provided some
illustrations of the latter in terms of applications, e.g., tracking, feature extraction,
segmentation, texture description, 3D geometry reconstruction, face recognition, and
fingerprint recognition. In many of these applications humans are experts, almost by
birth, and yet we do not have detailed knowledge on how this amazing performance
is achieved. Biased by this performance, the uninitiated can therefore not be blamed
for underestimating the challenges of image analysis, whether the issues are to be
addressed by her/his own visual system or by a computer.
When a computing system has to solve the same vision problem for multiple ap-
plications in changing environments, the versatility and robustness demands quickly
become challenging. In this topic, we discussed certain image analysis concepts and
tools in an attempt to make them useful in many scenarios. Intelligent vision systems
in changing environments are increasingly in demand not only because humans are
increasingly online while moving, but also because we want mobile machines to be
online and autonomous. Just as a hunting man must carefully consider what tools to
carry, so must a vision system that must function in changing environments be selec-
tive in its choice of processing tools because of the extreme resource limitations the
mobility demands.
In the face of undisputable evidence being accumulated, there is hardly a need to
argue about the existence of sophisticated signal processing in human vision, solely
devoted to
directional processing
. Such a mechanism, which is the sine qua non of
the most advanced mobile organisms, including the human, hunting or not, is at first
counterintuitive due to its computational “heaviness”. This is because directional
processing amounts to increasing the dimension of the visual signal in a hefty man-
ner. It is almost an explosion of data that takes place—and that routinely, before
any simplification, e.g., decision, takes place. Motivated by its imposing presence in
biological vision, this topic attempts to simplify and unify the modern directional
signal processing. It attempted to reconcile some of the most utilized principles in
