Generation of groundtruth data for visual saliency experiments using image segmentation
Rajarshi Pal, Pabitra Mitra, Jayanta Mukhopadhyay · 2012
In evaluating a model of a system, one should have the ideal response of the system as a reference. For evaluation of visual saliency models, one requires to obtain salient locations/objects in an image according to human visual system. In many experiments, eye-trackers are used for this purpose. But this technology is unavailable to many researchers. This paper deals with an alternate method of collecting groundtruth data. Salient locations are identified using a hand-eye coordination based approach. It does not require any instruments other than the usual computing resources. Then image segment based analysis generates groundtruth data which serves the reference for evaluating visual saliency models. The generated groundtruth data fits to the size and shape of the underlying objects.