Gabor Filter Banks

Gabor Filter: The Gabor filter is linear filter. It is a harmonic function (sinusoidal function) multiplied with the a Gaussian envelope function.

Gabor = Harmonic function X 2D Gaussian envelope.

The harmonic function has two attributes the wavelength and phase and the Gaussian envelope has sigma_x and sigma_y and theta the orientation of the ellipsoidal shaped Gaussian. When constructing a filter bank we need to construct many Gabor filter of various wavelengths and orientations. The gabor filters can be generated by the code found in /usr/sci/crcnsdata/CRCNS/Synapses/Code/Matlab/textureClassification/gabor_fn.m
The below image are a sample of these filters.

The Gabor filter response looks like a edge detection filter. The below images are the original & Gabor filter bank response. The filter bank is constructed for 20 different orientations. The max response of a pixel for each filter bank is chosen as the filter bank response.



Synapse Status: As the kNN classifier continues to run, I am going through the various filtering approaches of texture classification as described in here. This blog talks about Gabor filters in general.