Correlation is very time consuming. It is efficient only when
the object of interest is made up of standard shapes. Moreover the standard shape should be of the same size as that of the
target. So in order for correlation to work well the prototype should be of the same size and shape as that of the target
and that the target has to be static. This is because correlation is performed block wise on the image. If the target is in
random motion, the block may not be able to capture the target at all. We all know that the objects in our surrounding need
not be of a regular shape or even be universally same. Hence one more method to solve this problem is to use the object to
be recognized itself as a prototype. This only solves the problem of not having a regularly shaped prototype predefined, more
over the prototypes can be acquired dynamically, but what about their orientation and types? Consider a chair for example;
it need not be same universally having 4 legs, and also in an upright position during recognition. Correlation thus fails
in one or the other way since it requires each and every aspect of the object to match with that of the prototype.
How does it
happen in our visual system? We make use of an abstract capacity in us, which helps us build the object somewhat
vaguely and gives us an idea of it in another orientation. This abstract
capacity in us is our knowledge. Our knowledge is able to do this because we are able to perceive the objects in 3D. We are
able to think and therefore need not
have a prototype in all possible angles, which is impossible. An abstract answer though.
It's all about "Divide and Rule, Combine and Head" policy that
always wins.
