Skip to main content
. Author manuscript; available in PMC: 2016 Jan 13.
Published in final edited form as: Vision Res. 2014 Aug 29;103:116–126. doi: 10.1016/j.visres.2014.08.012

Table 2.

Regression of log10-transformed high-frequency spectral power in orthogonal and parallel orientations with slope as the only parameter. Results for both datasets show slopes close to unity in the background and greater than unity (and higher than background) in the figure, with their confidence intervals (CIs) non-overlapping. This indicates higher oriented spectral power orthogonal to the boundary than parallel to it on the figure side.

slope(radians) CI (low) CI (high) R 2
BSDS300 Figure (orthogonal vs. parallel) 1.036 1.030 1.043 0.53
Ground (orthogonal vs. parallel) 0.998 0.993 1.004 0.71
LabelMe Figure (orthogonal vs. parallel) 1.0722 1.065 1.079 0.53
Ground (orthogonal vs. parallel) 1.006 0.995 1.006 0.57