This project will try to develop a neural system that can solve the shape from texture problem where 3D vision percepts are generated by observing texture in 2D images. For example the figure below shows a 2D polka-dot texture pattern that gives a 3D shape percept. We wish to have a neural system that can take such a 2D image and provide a 3D perceptual/neural representation (i.e. a neural network that can assign elements, or boundaries, of an image to particular depths).
Our modeling work will focus on two things: (1) simulating the qualitative 3D shape perception of surfaces covered by homogeneously distributed polka dot textures; and (2) simulating psychophysical data describing how texture variations affect 3D shape perception.
To address our first goal we intend to simulate the 3D shape perception of the following surfaces viewed via perspective projection:
Todd et al., 2004
We would also like to simulate the perception of some surfaces viewed via orthographic projection:
A third category of images we would like to simulate are real world images. The two images we will consider are:
To address our second aim of analysing how texture variations affect 3D shape perception we will consider stimuli like those shown below:
In the above figure the left image has texture elements that are isotropic (not elongated) and the resulting depth percept is pretty weak, while the right image has texture elements that are anisotropic (elongated) and we experience a strong sense of depth. Why is this so? We plan to develop model mechanisms that describe data like this.
