Computer Vision: My long-term goal is to design an image-understanding system, which
mimics human image-categorization and image-search performance. The overall
approach is motivated by a number of perceptual and cognitive studies, which is
summarized in a monograph I have written a few years ago. So far I have
developed a contour partitioning, description and grouping process, which was
already successfully applied in image classification, image retrieval, shape
classification and video genre classification. To improve the system, we now elaborate
the grouping and learning process. We are eager to apply the methodology to
other tasks, in which contour information is beneficial (object recognition,
video surveillance, medical imaging, face analysis, etc.)
image classification I my
early work
image classification II reaching
some benchmarks with contour groupings
image
classification [site] shows
results from image classification (categorization), searches and sorting.
image
retrieval our
results at ImageCLEF 2010.
video retrieval how
the methodology improves video indexing.
shape retrieval a
quick and efficient way to describe shapes.
COREL category
labels contains
info about the labeling of the COREL image classes.
[link to visual
search] discusses
search aspects (e.g., regions of interest).
Applied Visual Perception: Human-computer interaction at visual interfaces, such as the PC
monitor, is still relatively awkward. The interaction could be easily improved
if human gaze position were tracked and guided toward task-relevant locations (Barth et al 2006). To realize that in practice, it requires a good
concept of how to read the user’s attention and how to generate the
appropriate process of gaze guidance. I have assembled a list of aspects, which
allows me to systematically approach this issue, and I have carried out basic
eye-movement research to investigate some of these aspects in detail. Some of
my applied ideas were implemented in collaboration with Michael Dorr at the
[link to gaze
guidance] surveys
my part of the research collaboration.
[link to
applied project] a
gaze-recapturing editor cursor (GREC) to improve text-editing on a PC monitor.
[link to
saccadic target selection] an analysis of saccadic target
selection.
[link to project
website] 5
different European labs were involved.