| DIFFUSED AND FOCUSED ATTENTION |
Hezy Yeshurun
Dept. of CS, Tel Aviv University- Israel
e-mail: hezy@math.tau.ac.il
Computerized systems use "attentional" algorithms to detect areas
where resources should be directed. These methods, referred to as
Detection of Regions of Interest in Computer Vision are usually based
on edge maps.
However, edge maps might fail to convey all necessary information for
this task, especially in complex scenes (e.g. noisy, textured or cluttered
images).
We suggest a novel non-edge-based mechanism for
detection of areas of interest in images, which extracts three dimensional
information from the image. The operator assumes no a priori knowledge on
the image, and thus can be invoked in an early stage of the processing.
Our operator detects smooth convex and concave objects based on direct
processing of intensity values. Invariance to a large family of
functions is mathematically
proved. It follows that our operator is robust to illumination
variations, and to variations in scale and orientation, in contrast with most
other attentional operators which
demand an a-priori knowledge of the scale. The operator is also demonstrated
to efficiently detect 3D objects camouflaged in a noisy area.
An extensive comparison with
edge-based attentional operators is provided.
e-mail: hmp@ipvvis.unipv.it