Modeling feature-integration in human vision with drift diffusion models

Abstract:
The world surrounding us is inherently noisy and so is the input into the visual system. Hence, the visual system is constantly involved in a perceptual decision making process. It has to decide, what the most likely event was, that caused the input. The reaction times of such perceptual decisions are generally modeled by drift diffusion models, in which sensory input is accumulated across time until a threshold is reached and thereby a decision is taken. Most studies in this field have used a constant sensory input and a constant drift value of the diffusion process. Using such constant input designs, the possibilities to investigate the relationship between sensory input and the diffusion's drift are thus very limited.
Here we use a feature fusion paradigm, to investigate how time varying input is accumulated in integration models. In feature fusion, features of stimuli, presented in rapid succession, fuse into one percept. E.g. a yellow disk is perceived, when a red disk is followed by a green disk. By using such time varying input we show that the integration is not driven by the input directly but by the output of a preliminary sensory integration process.

back