Variations of Gabor - Wide Vs Narrow Gaussian Envelopes

Here, we investigate the effects of varying one of the parameters to the family of Gabor masks. Although the variation is straightforward, it has considerable implications for the theoretical underpinnings of the Gabor Energy method.

We implement the Wide mask with standard deviation of the Gaussian equal to one quarter of the mask width. This will be approximately 4 pixels for these tests since a mask width of 17 pixels is used throughout.

We implement the Narrow mask by fixing the standard deviation of the Gaussian proportional to the wavelength~; specifically, equal to one half the wavelength.

The Table below shows the per-test-suite scores for the Narrow and Wide Gabor masks. The Narrow masks performed best on the microtextures, and the Wide masks performed best on the macrotextures.