Fast Computation of the Gabor Wavelet Transform
Gareth Loy · 2002
The Gabor filter is a valuable tool in computer vision, however, its high computational load precludes its use in many applications. This paper presents a novel method for approximating Gabor wavelets with a function that can be convolved more efficiently in the spatial domain. The method extends the well known separable nature of the Ga-bor filter, and demonstrates a means for reducing the com-putation required when determining the separable 1D con-volutions. Experimental results are presented showing the performance of the new method. It is demonstrated that the new method provides an accurate approximation to the con-volution with the 1D separated wavelet component of the Gabor kernel, while requiring only 60 % of the computation normally required for this operation. 1