A computer vision system gives special importance to pattern recognition. Generally
there are some known patterns that are stored in the database and, a matching and grouping technique is used to find out a
pattern. Matching is done using correlation. Why can't correlation solve shape recognition? How does a vision system know what a pattern is? When we are born our brain does not contain anything
other than the BIOS. More over it does not know any language. So what does it understand when it sees the objects around it
or someone speaks of something. Whatever the object may be, for it, its just a colored patch. Then how does it start learning?
They tend
to get attracted towards bright colors and movement. This is because of the interrupt I was talking about under the motion heading. This is the point from where the process
of seeing and learning starts. We tend to observe things that draw our attention and segment the possible ones. Segmentation is a technique of categorizing the objects in
the scene.
Grouping together, objects of similar
color can do segmentation. At the first glance the objects of similar color are segmented. When one goes to observe the region of a particular
color the precession increases and we observe its various shades. Since segmentation
is basically based on differences in color as one starts to observe minor variations
in color he will be able to segment smaller and smaller objects in the region of the fovea. Similar segments join hands to give you a pattern.
Shape is a combination of segment along with depth. Shapes
can be 2D or 3D. A single object may have different colors, so if segmentation alone is used definitely it will break the
object into pieces. Then how do we perceive shape? This is where continuity asks a role.
|
|
