Neurodynamical evidence of gaze prediction decrease with saccade number

Previous psychophysical research show that saliency is elucidated during the first and second saccade. Latest research using computational models for saliency prediction promote increasing patterns of  prediction error in saccade landing and saccade amplitude with respect to the saccade number, suggesting the scepticism of including specific fixations temporally on the evaluation of saliency metrics using real image datasets. Acknowledging the outstanding results on saliency prediction from such models using temporally-invariant metrics, our evidence suggests that a gaze-based framework might help to minimize prediction error on both scanpath and saliency metrics. In order to evaluate such phenomenon, we have modeled a neurodynamical recurrent network using an excitatory-inhibitory firing-rate model of V1’s lateral interactions, being responsible of bottom-up visual saliency. Our model provides foveation by including the cortical magnification towards V1’s retinotopy. We show the results suggested from scanpath prediction and we also provide a comparison of saliency prediction metrics from other computational models for further analysis. Moreover, we aim to highlight the biological plausibility of our model previously presented to work using the same parameters for brightness induction and simultaneously reproducing perceptual processes such as chromatic induction, binocular disparity and visual discomfort.