Artificial Intelligence and Soft Computing: Behavioral ... - Arteimi.info

co-ordinate systems. The correspondence between each two points on
different images is then determined. One simple way to handle the
correspondence problem is to first represent all 2-D points into 3-D points and
then identify which point in one image has a close resemblance in another
image. There are many other approaches for multi-sensory data fusion.
Dempster-Shafer (D-S) theory, for instance, is one significant tool to handle
this problem. The mapping of real image data onto basic probability
assignments in D-S theory, however, is a problem of practical interest. Neural
tools can also be used for eliminating uncertainty in multi-sensory data, but
here too mapping from image domain to neural topology is a practical
problem.
17.5 Conclusions
Vision systems can be functionally classified into three main levels, namely,
low, medium and high level vision. AI is mostly used in the high level vision.
The high level vision mainly deals with recognition and interpretation of 3-D
objects from their 2-D images. There exist many approaches to interpret a
scene from more than one image. Kalman filtering is one of such techniques.
Its main advantage is that it employs a recursive algorithm and thus can
update an estimator from input data in an incremental fashion. The vertices of
points in a 2-D image can be first mapped to their 3-D locations by
supplying the same 2-D points from multiple images. Now, we can construct
the equation of 3-D lines from 3-D points by a second stage of Kalman
filtering. Lastly, we can determine the equation of the planes containing more
than one line. The spatial relationships among the planes are then analyzed to
determine the 3-D planer object.
Exercises
1. Draw on a graph paper a 2 level (binary) image and verify the edge
detection algorithms manually. Note that here one pixel is equal to one
smallest cell on the graph paper.
2. Write a program to find edges from a given (64 x 64) gray image using
Sobel masks.
3. Wrte a program to verify the gaussian filter.
4. Can you derive the gaussian filter mask from the supplied filter
equations?