Compatibility of a competition model for explaining eyefixation durations during free viewing

Abstract

Intersaccadic times or eye fixation durations (EFD) are relatively stable at around 250ms, equivalent to 4 saccades by second. However, the mean and standard deviation are not sufficient to describe the frequency histogram distribution of EFD. The exgaussian has been proposed for fitting the EFD histograms. Present report tries to adjust a competition model (C model) between the saccadic and the fixation network to the EFD histograms. This model is at a rather conceptual level (computational level in Marrs classification). Both models were adjusted to EFD from an open database with data of 179473 eye fixations. The C model showed to be able, along with exgaussian model, to be compatible for explaining the EFD distributions. The two parameters of the C model can be ascribed to (i) a refractory period for new saccades modeled by a sigmoid equation (A parameter), while (ii) the ps parameter would be related to the continuous competition between the saccadic network related to the saliency map and the eye fixation network, and would be modeled through a geometric probability density function. The model suggests that competition between neural networks would be an organizational property of brain neural networks to facilitate the decision process for action and perception. In the visual scene scanning the C model dynamic justifies the early post-saccadic stability of the foveated image, and the subsequent exploration of a broad space in the observed image. Code to extract the data and to run the model is added at the supplementary material.