Rotation- and scale-invariant texture features based on spectral moment invariants

Area of Science:

• Image processing and computer vision
• Pattern recognition
• Fourier analysis

Background:

• Traditional texture analysis methods often struggle with variations in scale and orientation.
• Existing Fourier transform domain methods typically rely on ad hoc measurements of power spectrum shape.
• There is a need for robust texture feature extraction methods invariant to geometric transformations.

Purpose of the Study:

• To extend moment invariants to the Fourier transform domain for texture analysis.
• To introduce spectral moment invariants (SMIs) for quantifying texture signatures in image power spectra.
• To develop a method for extracting rotation- and scale-invariant texture features directly from the Fourier spectrum.

Main Methods:

• Applied established moment invariants to the power spectrum of images.
• Defined and computed spectral moment invariants (SMIs) using complex spectral moments.
• Utilized SMIs for texture classification experiments to evaluate discriminative capability.

Main Results:

• SMIs systematically extract rotation- and scale-invariant texture features.
• The proposed SMI method directly quantifies texture information from the Fourier spectrum.
• Texture classification using SMIs achieved performance comparable to or better than spatial or wavelet domain methods.

Conclusions:

• Spectral moment invariants provide a powerful new tool for invariant texture feature extraction.
• This method offers a direct and systematic approach to texture analysis in the Fourier domain.
• SMIs show significant potential for improving texture recognition and classification accuracy.