Skip to main content
Figure 11 | BMC Neuroscience

Figure 11

From: A neural computational model for bottom-up attention with invariant and overcomplete representation

Figure 11

Overcomplete bases obtained by PCICA. A clear topography emerges from the two maps. Though both sets consist of Gabor-like filters, filter patterns and distributions learned from different datasets appear to be different, particularly in their average length and frequency. (a) the basis set that is used in most experiments (except the experiment on Weizmann dataset). It is learned from natural images. An example showing the filters in neighborhood slowly change their properties as marked within a red square; (b) the basis set that is learned from Weizmann dataset. Some bases are similar to receptive fields of curvature-selective, end-inhibition, side-inhibition and texture-selective cells.

Back to article page