Robustness of shape descriptors and dynamics of learning vector quantization

In this thesis the author deals with two important aspects of
a computer-based pattern analysis system, viz., representation and classification.
The representation involves extraction of
characteristic features of the patterns and the classification deals with learning and
discriminant analysis. The author investigates the performance of:
(a) representation schemes of
contour based shape patterns of objects present in digital images and (b)
Learning Vector
Quantization (LVQ) classifier in a proto-typical scenario. In the context of (a)
more specifically the robustness of contour based shape descriptors to incomplete
representations of objects is investigated. Using concepts of statistical physics
and theory of on-line learning, the dynamics and the generalization ability of
LVQ algorithms in a given data model are studied in the context of (b).

	Title and contents
	Acknowledgments
	Chapter 1
	Chapter 2
	Chapter 3
	Chapter 4
	Chapter 5
	Publications
	Samenvatting
	Bibliography
	Index
	Complete thesis

Meer informatie in de catalogus
Meer informatie in Picarta

Afdrukken op bestelling.