Localization of Salient Objects in Scenes through Visual Attention

Abstract

There are several real situations in which it is useful to have a system able to detect a salient object and its localization in a given scene in autonomous way. Among these systems are robot vision systems that must detect determined objects, security systems, that must detect stranger people or objects in environment. These systems need to present a very fast performance and accuracy. In this article, we have combined two neural networks, a Kohonen model and a Pulsed Neural Network for the creation of an attributed-saliency map. Thanks to this combination, it is possible not only to detect the salient object in given scene as well as its localization in the image. Several tests have been performed to verify the viability of the model as a mechanism of selection of objects as a part of a visual attention system. The results demonstrate that the technique proposed is very fast for detecting the region of the image corresponding to salient object.