Back to search

National Technical University of Athens (NTUA)

Αλγοριθμικές μέθοδοι επεξεργασίας-ανάλυσης πολυμεσικής πληροφορίας

Abstract

dc:description

The present thesis proposes novel techniques for intelligent signal processing giving emphasis on image and image sequence segmentation and pattern recognition. By introducing high resolution images and video frames, the traditional processing has to deal with an all the more larger pixel space. Especially in the case of image segmentation and pattern recognition, the computational cost is very high, rendering the techniques unsuitable for many applications. This thesis proposes the Operator Context Scanning algorithm to deal with the aforementioned problem. This algorithm uses sliding window operators introducing an Additive Increase/Multiplicative Decrease velocity scheme. The algorithm focuses only on regions of interest, that is, the regions that expose the operator characteristics, overlooking a large part of the pixel space of an input image or frame. As a result, the computational cost lowers significantly and scales well as the input image resolution increases. Furthermore, a novel image characteristic is proposed, namely the image context based on an operator, which derives from the operator context scanning algorithm, in order to be used as a fast means of categorizing images. Next, two systems are presented that implement the proposed scanning algorithm, namely, the license plate recognition system and the video matching system of television broadcast advertisements. According to the first, the proposed algorithm is used to segment the license plate region from the input vehicle image. Compared to the traditional exhaustive pixel space scanning, the algorithm processes only a very small portion of the input image to effectively segment the license plate region. The second system follows the principles of the Operator Context Scanning algorithm to process and match video sequences. In this case, the algorithm is used as an alternative to the traditional block based processing to quickly calculate Regions of Interest that include high density operator characteristics. Then, these regions are compared to measure the similarity of TV stream video sequences and advertisements. In the final part of this thesis, the application of machine learning techniques and their ensembles are presented as a means of pattern recognition in different scientific fields. Initially, machine learning based methods are implemented on the field of electronic learning in order to predict student achievement. Then, a machine learning ensemble is proposed for character recognition. Finally, the ensemble is used to process a data sample that includes patients with Alzheimer’s disease.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2010

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Giannoukos, Ioannis
  • Γιαννούκος, Ιωάννης

Subjects

dc:subject × 14

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/22949

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

Giannoukos, Ioannis; Γιαννούκος, Ιωάννης. Αλγοριθμικές μέθοδοι επεξεργασίας-ανάλυσης πολυμεσικής πληροφορίας. National Technical University of Athens (NTUA), 2010. http://hdl.handle.net/10442/hedi/22949