Detecting distinguished regions by saliency

Abstract

A method for detecting and characterizing local image regions based on saliency is introduced. The proposed method detects scale localized salient regions in an image by a saliency operator which uses the concept of visual attention. A new descriptor based on a cornerness measure is presented which allows a stable identification of regions of interest and at the same time allows for an elaborate description of the identified salient regions. Experiments demonstrate that the resulting salient regions and their descriptions are discriminative enough for image matching. © Springer-Verlag; 2003.