A fundamental goal of vision research is to understand how vision works in natural conditions. Vision systems are matched to the critical tasks that organisms perform to survive and reproduce. Thus, it is fundamentally important to analyze vision systems with respect to these tasks, and the properties of natural stimuli that are relevant to those tasks. My lab takes the following approach. First, we measure task-relevant statistical properties of natural stimuli. Next, given biological constraints, we determine how to optimally use those properties to perform the tasks. Then, we formulate hypotheses based on the first two steps and test them in behavioral experiments with natural stimuli. To connect our results with the classic literature and determine the generality of our results, we also collect data with artificial stimuli. Using a unique suite of natural image databases, computational tools, and psychophysical paradigms (many of which have been developed or refined in our laboratory), we propose to investigate several fundamental tasks relevant for the estimation of depth and motion in natural scenes.
Aim 1 investigates optimal and human disparity estimation in natural stereo-images.
Aim 2 investigates optimal and human motion estimation in natural image movies.
Aim 3 investigates optimal and human motion-in-depth estimation in natural stereo-image movies. Many of the proposed studies will be the first to characterize the statistical properties of natural images that underlie the human ability to perform these tasks accurately. Many of the proposed studies will also be the first to measure human performance in these tasks using natural stimuli. The result of these studies will be not only unique new measurements, but new principled models that can predict human performance under natural conditions and guide future behavioral and neurophysiological studies of the underlying mechanisms. Encouraging preliminary results have been obtained for many of the proposed studies. !
A fundamental goal of vision science is to understand how vision works in the real world. This proposal is directed at understanding the human ability to estimate and discriminate depth and motion in natural scenes from images formed on the retinas. We will measure task-relevant natural image statistics, and develop ideal observer models that specify the optimal computations given known biological constraints. These models provide rigorous hypotheses against which to test human performance in behavioral experiments. Progress in understanding natural vision will advance basic science and may yield useful technological advances. !
|Iyer, Arvind V; Burge, Johannes (2018) Depth variation and stereo processing tasks in natural scenes. J Vis 18:4|