The visual brain faces the difficult task of reconstructing a three-dimensional (3D) world from two-dimensional (2D) images projected onto the two retinae. In doing so, visual information is organized in terms of objects in 3D space, and this organization is the basis for visual perception. In complex visual scenes, both the foreground and the background are rich in features of different types. The brain must find a way to group together the features that belong to objects on the foreground and distinguish them from features in the background. The goal of this thesis is to understand how the neural circuits in primate cortex accomplish this task using grouping mechanisms for object-based vision and attention. Through computational modeling, I show that grouping mechanisms are fundamental for linking early feature representations to tentative perceptual objects known as proto-objects. Previous models on the neural coding of border ownership have identified a plausible network architecture for proto-object based perceptual organization. I extend these models to explain how the same grouping model framework can be used to perform contour integration, border-ownership assignment, grouping of 3D surfaces, and computation of 3D visual saliency. My models offer several falsifiable predictions which can be tested in future experiments. My models also clarify how top-down attention interfaces with the neural circuits responsible for grouping together the features of an object. Overall, the models developed address the important question of how visual features are grouped into 2D and 3D object representations.
【 预 览 】
附件列表
Files
Size
Format
View
Grouping mechanisms for object-based vision and attention
PDF
download
878987797
028-85220240
