The talk's first part will present an extension to global similarity
measures on polygonal boundaries to discover partial matches.
The presented partial similarity measure emphasizes the common
parts of objects compared to those distinguishing them.
Clearly such a measure can be utilized as the core of shape based
image retrieval, but apart from that it offers solutions to many
kinds of problems in applied computer vision. The talk will give an
example how partial shape similariy offers a solution to a main problem in
robotics: mapping of environment data gained by a laser range scanner.

Along with the application of robot mapping, the talk's second part will
go the opposite direction (from robotics to object recognition): presenting a modified 
version of EM (Expectation Maximization),the modification tailoring the classical EM approach to
robot mapping, the connection to EM based edge finding in images and grouping edges
to object parts will be made. These parts can be the input to the aforementioned
partial shape similariy measure in 2D. But the modified EM algorithm also offers
interesting object decompositions into parts of 3D objects, a decomposition
that offers a base for 3D similarity measures. The talk will present examples of
3D decompositions and show properties and behaviour of the algorithm to 
motivate possible 3D similarity measures.