Investigating the effect of texture orientation on shape perception
We are interested in investigating the effect of texture orientation
on shape perception, with the goal of determining how to use texture to
enhance the representation of surface shape for visualization applications.
To this end we have developed an efficient algorithm for synthesizing a high
resolution texture pattern (given by a 2D sample image) over an arbitrary
polygonal surface in such a way that the orientation of the texture follows
a specified vector field at a per-pixel level. We are in the process of
conducting a series of experiments to evaluate the effects of various texture
conditions on various kinds of shape judgements.
Texture orientation follows a constant "up" direction
Texture orientation follows the direction of greatest normal curvature
Texture orientation follows the direction of least normal curvature
Left: texture orientation follows the direction of least normal curvature
Right: texture orientation follows a constant "up" direction
These are examples of some of the sample textures and textured surfaces that
we have used in our latest experiments investigating the effect of various
texture characteristics on shape perception. The task for the observer is to
indicate the surface normal direction using a Koenderink-style probe (not
shown in these images).
From left to right: two-directional line texture, one-directional line texture,
one-directional lic texture samples.
A surface with the two-directional line texture, oriented to follow both of the
The same surface with the one-directional line texture, oriented to follow the
first principal direction only. Notice the abrupt pattern discontinuities that
occur where the first and second principal directions "switch places".
The same surface with the one-directional lic texture, oriented to follow the
first principal direction only. Texture discontinuities appear less prominent
with this more complex pattern.
The same surface without any texture.
We are currently running studies that look at the effect of global vs
local information (using a window to hide global information), the effect
of perspective vs orthographic projection, and the effect of using ceiling
vs ground plane based surfaces.
Here are some pictures from an earlier experiment, in which we used
textures synthesized via line integral convolution. These textures
were very easy to create but they have the disadvantage that they can
only represent one of the two principal directions at a time.
The paper (presented at Human Vision and Electronic Imaging VI) is
Texture orientation follows a vector field of random directions
Texture orientation follows a vector field of uniform (constant) directions
Texture orientation follows a vector field of coherently varying directions
Texture orientation follows a vector field of first principal directions
Poster presented at Vision Sciences Society meeting 2002