A Visual Attention Model for Omnidirectional Images
Amirhossein Habibian, Ali Borji, Majid Nili Ahmadabadi, Babak Nadjar Araabi · International Conference on Intelligent Information Processing · 2010
Visual attention is one of the mechanisms existing in the brain of human and other primates that directs their perception to limited regions of the observed scenes which are probably more important and relevant. Inspired by this, several computational models have been proposed and are used in many computer vision and robotic applications. Since these models are proposed for conventional images, they are not suitable for being used for omnidirectional images which suffer from large amounts of non linear deformations and distortions. In this paper we propose a model of visual attention which uses some properties of omnidirectional images to modify the saliency- based model of visual attention for these kinds of images. This model attends to the regions which are not only salient but also placed on the image’s radial lines, and so, are invariant to the deformations. In some experiments we show our proposed model has better performance than saliency-based model of visual attention for the task of omnidirectional image classification.