Saliency map estimation by constructing graphs of possible eye-tracking paths

We introduce a novel approach for saliency detection where we fuse perceptional saliency with machine saliency in a statistical approach. The improvement of our fused algorithm against other methods is presented. Human saliency is recorded from human eye movement during free view training. The transition movements caused by the saccades are evaluated to generate transition probability tables. This new kind of training is applied into a connection graph based model where transitions among the machine generated saliency points are weighted by the probabilities derived from the human trained probability table. In the presented method, different psychophysical studies are taken into consideration by inferring regions of interests. The proposed method results in a good estimation of the possible interest areas of the human vision measurements.