SEMINARIOS DE INVESTIGACIÓN EN INGENIERÍA INFORMÁTICA Y DE
TELECOMUNICACIÓN 2007-2008
Actividad de Formación Continua del Programa Oficial de Posgrado en Ingeniería
Informática y de Telecomunicación
Escuela Politécnica Superior, Universidad Autónoma de
Madrid
Lunes, 9 de Junio de 2008, 12:00
Salón de grados, Escuela Politécnica Superior,
Universidad Autónoma de Madrid
Region-based
segmentation utilizing multiple sources of information, its applications and
evaluation
Tomasz Adamek
Telefónica I+D, Barcelona
Este seminario está parcialmente financiado por el
Programa ProMultidis (S-0505/TIC-0223)
de la Comunidad Autónoma de Madrid
Abstract
The
talk starts by discussing the importance of image segmentation in the context of
content-based image retrieval, outlining the main categories of approaches, and
briefly reviewing the most characteristic segmentation methods. This is
followed by a presentation of a new method for segmentation of images into large
regions that reflect the real world objects present in the scene. Several
extensions to the well-known Recursive Shortest Spanning Tree (RSST) algorithm
are discussed among which the most important is a novel framework for
integration of evidence from multiple sources with the region merging process
based on the Dempster-Shafer (DS) theory that allows integration of sources
providing evidence with different accuracy and reliability. Other extensions
include a new color model and colour homogeneity criteria, practical solutions
to analysis of region’s geometric properties and their spatial configuration,
and a new simple stopping criterion aimed at producing partitions containing the
most salient objects present in the scene. Extensive experiments indicate that
the proposed solutions improve the correspondence of the segmentation to
semantic objects present in the scene. The proposed method was used by the K-Space
project for participation in the TRECVid'07 evaluation activity and for
improving usability of supervised image segmentation. Finally, the subject of
objective evaluation of segmentation quality is discussed in more detail as it
has received relatively little attention in the literature, especially when
compared to the large number of publications on the topic of image segmentation
itself.
Tomasz Adamek
Tomasz Adamek received the M.Sc.
degree in Telecommunications from Poznan University of Technology, Poznan,
Poland in 2001 with the highest honour and the Ph.D. degree in Electronic
Engineering from Dublin City University, Dublin, Ireland in 2006. Between 2006
and 2007 he worked as a post-doctoral researcher at Centre for Digital Video
Processing, Dublin City University, Ireland. Currently he is working as a
technology specialist in the area of analysis and classification of images at
Telefónica Research and Development. His research interests include shape
analysis, region-based and object-based segmentation, and video sequence
analysis for indexing applications. In this area he has authored and co-authored
over 20 academic papers in high-impact journals and conferences. He has
participated in several EU funded projects, such as SCHEMA, aceMedia, and K-Space.
Specifically, he was involved in coordinating research activities in the area of
Moving 2D and 3D Object Segmentation and Indexing and in Designing and
Instantiating the Technical Infrastructure required to enable and facilitate
integration activities within the K-Space project.