Segmentation Of Moving Images By The Primate Visual System PDF Download

A fundamental goal of sensory processing is predicting the future state of the environment. Visual prediction is difficult because the stream of images on the retina evolves according to irregular, curved trajectories. We posit that the visual system seeks to transform natural inputs such that neural representations follow straighter and more predictable trajectories. Previous work has provided psychophysical evidence for this hypothesis by showing that the human visual system selectively straightens the temporal trajectories of natural movies, thus facilitating their extrapolation. In this thesis, I investigate the neural basis of perceptual straightening. We hypothesize that perceptual straightening is achieved through a series of cascaded transformations and found supporting evidence from psychophysical and neurophysiological measurements. We estimated the curvature of internal trajectories from human perceptual judgements across distinct sub-regions of the visual field. Perceptual results suggest that computations that underlie object recognition overlap with those that straighten natural movies, thereby revealing deficits in both processes--crowding and absence of straightening in the periphery. We posit that straightening occurs at multiple stages in the ventral stream as it is linked to recognizing objects. Next, we investigated if straightening occurs as early as the primary visual cortex (V1). We developed a novel computational tool to infer trajectories from neural populations and found robust straightening in V1 recordings from anesthetized macaques. Next, we explored the computational basis of V1 results with an image-computable model. We built a model to describe well-known computations of V1 cells and tested the model with the same movie stimuli. Model-predictions reveal that these computations only partially engage the mechanisms for straightening, revealing the involvement of more complex visual mechanisms. To further test our hypothesis, we asked if additional straightening occurs across the V1-V2 cascade. We recorded populations in areas V1 and V2 from awake fixating monkeys. Consistent with our hypothesis, we find incremental straightening between V1 and V2 elicited by the same natural movies. Together, we find collective evidence of neural straightening as an emergent property that contributes to perceptual straightening in a progressive manner starting from the early stages of vision