Representation and recognition

To recognise individual objects in a digital image, there are two problems that need to be solved:

• given a collection of named objects, construct a computer representation or model of each object, and
• given a digitized sensor image (intensity or range data, for example), find either or all of the following:

  1.

  an identity or name,

  2.

  the location, and

  3.

  the orientation

  of each object in the scene.

An example occurs in the polyhedral world where the task is to match predicted and observed line labelings through a graph isomorphism or endomorphism search. We will consider this in more detail later, but will first concentrate on some of the common modelling schemes used in computer vision.

 

Figure: From image capture of the real world to object recognition.