Back to search

National Technical University of Athens (NTUA)

Αναπαράσταση σχήματος, ταξινόμηση και ανάκτηση εικόνων

Abstract

dc:description

The Dissertation is focused on issues of representation, coding,classification and retrieval of visual information either in the form of shape orin the form of pictures in an approach that is compatible with biological butalso functional data concerning human perception.In the case of shape representation and coding notions of biological butalso intuitive compatibility are introduced. The new concept of “curvatureskeleton” is presented as a new shape representation technique that iscompatible with the Human Visual System. The “Curvature Skeleton” is aglobal shape descriptor that also contains curvature measurements of theshape’s contour, embedded in the gray scale intensities of the skeleton image.The method is inspired and thus compatible with the Human Visual System andat the same time improves the recognition capability of Neural Networkclassifiers in experiments of shape classification under conditions of occlusionand deformation. Furthermore, a new transformation, the “Global-LocalTransformation” is also presented. It is shown that this transformation providesa framework for defining various descriptors of shape that offer newpossibilities in invariant shape recognition and is also compatible with thehuman intuition of shape. A mathematical analysis based on differentialgeometry of planar curves, leads to the “Global-Local Equations”, a closedform relation between global, local and hybrid shape visual features. Theresulting representation is shown to be particularly effective in the presence ofnoise. In various experiments, an improvement in the complexity but also inthe precision\recall score of widely used and state of the art methods isachieved.In the case of indexing and retrieval of pictures, novel knowledgerepresentation techniques are introduced in an approach to link to task of imageunderstanding to that of representing and exploiting knowledge about thisimage. A new method for knowledge representation with regard to indexingand retrieval of images in information access systems, like Google ImageSearch, is presented. The “Markovian Semantic Indexing”, provides a newway of assigning semantic interconnections between keywords that are linkedto the images by means of an interactive procedure between the system and theusers that retrieve these images. A distance between the keywords isconstructed based on a Markovian model of general user behavior. The samemodel is used to produce a vector representation of the chosen by the usersimages and annotate them in an automatic manner. The annotations, attributedto the pictures automatically by the system are then used in a Markoviancontext to index the pictures. Related work is examined and with experimentsis shown that the proposed method performs better when compared to LatentSemantic Indexing in the case of Annotation Based Image Retrieval.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2011

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Raftopoulos, Konstantinos
  • Ραυτόπουλος, Κωνσταντίνος

Subjects

dc:subject × 20

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/33249

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Raftopoulos, Konstantinos; Ραυτόπουλος, Κωνσταντίνος. Αναπαράσταση σχήματος, ταξινόμηση και ανάκτηση εικόνων. National Technical University of Athens (NTUA), 2011. http://hdl.handle.net/10442/hedi/33249