• What?

To recognize and identify a 3D object from an input 2D image.

• How?

By moving the representation of the 3D objects to a set of 2D images, the matching complexity is greatly reduced.

• But...

How many images needed to represent a 3D object?
• Too Many?

Disk constraints can be impractical.  Too many comparisons to database causes slow results.
• Too Few?

Shape is ambiguous, i.e. bad recognition performance.
 

There is a tradeoff between quality of representation and time/space constraints/performance.

So, how do we determine the optimal number of images?

Next:  Page2