A zernike moment phase-based descriptor for local image representation and matching

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Robust local image descriptors are essential for computer vision tasks.
• Existing descriptors face challenges with photometric and geometric transformations.

Purpose of the Study:

• To evaluate region descriptors for image understanding.
• To propose and validate a novel Zernike moment (ZM) phase-based image descriptor.

Main Methods:

• Normalized circular regions derived from elliptical interest points.
• Representation using Zernike moment phase information.
• Comparison with SIFT, GLOH, PCA-SIFT, complex moments, and steerable filters using precision-recall.

Main Results:

• The proposed ZM phase descriptor demonstrates superior performance across common transformations.
• Accurate rotation angle estimation is a key factor in its effectiveness.
• The descriptor shows robustness to non-uniform intensity fluctuations.

Conclusions:

• The ZM phase descriptor offers a highly effective solution for robust region matching.
• Feature dimensionality, orthogonality, and rotation estimation accuracy influence descriptor performance.
• This descriptor advances image understanding applications requiring transformation invariance.