Turker Kuyel and Joydeep Ghosh,
A fast space localized computation of the outputs of a
Gabor filter bank,
In Proceedings of the IASTED Conference on Signal and Image Processing (SIP'95),
Las Vegas, USA, pp. 511-514, November 1995.
Turker Kuyel and Joydeep Ghosh,
A fast space localized computation of the outputs of a
Gabor filter bank,
In Proceedings of the IASTED Conference on Signal and Image Processing (SIP'95),
Las Vegas, USA, pp. 511-514, November 1995.Gabor filters have been widely used for modeling cortical cell frequency and orientation selectivity properties, as well as for image processing routines including segmentation and shape-from-texture [1], [2], [3]. Because there are many frequency and orientation selective channels in the early visual cortex, the computation of multiple Gabor filter outputs are necessary for the computational modeling of the human visual cortex. However, obtaining the outputs of these frequency and space localized Gabor filters is computationally very expensive. A local Fourier transform based technique is proposed to compute the outputs of multiple Gabor filters. Significant speedup, approaching one order of magnitude is obtained over the standard FFT based convolution for the computation of multiple Gabor filter outputs. This algorithm saved many hours of processing time in the simulations of the human cortical processing we have been performing on the massively parallel computer, the MasPar MP-1104.