Fast 2D Complex Gabor Filter With Kernel Decomposition.

Area of Science:

• Computer Vision
• Image Processing

Background:

• 2D complex Gabor filtering is widely used in computer vision and image processing.
• Computing Gabor filter banks at multiple orientations and frequencies is computationally intensive with existing methods.

Purpose of the Study:

• To propose an efficient method for computing 2D complex Gabor filter banks.
• To reduce computational redundancy in multi-orientation and multi-frequency Gabor filtering.

Main Methods:

• Decomposition of the Gabor kernel for separable convolution with Gaussian kernels.
• Reusing intermediate results computed at specific orientations to accelerate filter bank generation.
• Extension to a fast 2D localized sliding discrete Fourier transform using Gaussian kernels.

Main Results:

• The proposed method significantly reduces runtime for Gabor filter bank computation.
• Achieved faster performance compared to state-of-the-art methods.
• Maintained filtering quality comparable to existing techniques.

Conclusions:

• The novel approach offers an efficient solution for generating 2D complex Gabor filter banks.
• The method optimizes computational redundancy, leading to faster image processing.
• This work advances efficient Gabor filtering techniques for computer vision applications.