An approach to automatic detection and extraction of regions of interest in still images

Abstract

Localisation and extraction of a salient region in an image constitutes an important task in computer vision. It is a very challenging problem due to its ill-posed nature. Furthermore, it is not always evident even for humans. In this paper we present an approach for automatic detection and extraction of salient regions. The main core of the algorithm is based on Ullman and Koch's model of the bottom-up attention, however their approach usually predicts only the place of a salient region. Therefore, in order to improve the achieved results a segmentation method based on k-means has been adopted. In addition, a few heuristic limitations are applied to the previous segmentation method in order to prevent it from creating inadequate segments. We show that our approach can render quite precise results in localisation and extraction tasks, taking into account the human point of view. © 2013 Springer-Verlag.