se

pro is ov p sk-d pos of ed w wi ual fea movem s also also i ne tha gaged r a sp of Yarbus using new equipment and stimuli, and with larger num-bers of subjects. For instance, in Tatler et al. (2010) the results obtained by Yarbus were confirmed in an experiment that studied the effect of instructions in viewing a portrait of Yarbus. While the effect of visual-task on eye movement pattern has been thoroughly e inverse Yarbus e learning and testin ach perfor specific visual task on an image, and extracted a feature from the eye movement records. The feature vector used set of seven summary statistics of eye movements, which are often used in scanpath analysis (Castelhano & Henderson, 2008; Mika et al., 1999). This feature vector included, among others, the num-ber of fixations, the mean fixation duration, the mean saccade amplitude and the portion of the image covered by fixations. The machine learning approaches used three different classifiers based on linear discriminant analysis (Mika et al., 1999), correlational ⇑ Corresponding author