Skewed rotation symmetry group detection

Area of Science:

• Computer Vision
• Image Analysis
• Computational Geometry

Background:

• Symmetry detection is crucial for image understanding and analysis.
• Existing methods struggle with affinely skewed and multiple rotation symmetries.

Purpose of the Study:

• To present a novel algorithm for detecting affinely skewed rotation symmetry groups in real-world images.
• To define and detect five key properties of skewed rotation symmetry groups.

Main Methods:

• Proposed a frieze-expansion (FE) method to simplify rotation symmetry detection to 1D translation symmetry detection.
• Utilized rotational symmetry saliency maps and local features.
• Applied Discrete Fourier Transform (DFT) frequency analysis on frieze-expansion patterns (FEPs).
• Employed FEP phase information for rectifying affinely skewed symmetries.

Main Results:

• Successfully detected multiple rotation symmetry groups, including their types (cyclic, dihedral, O2), cardinalities, and supporting regions.
• Demonstrated superior performance over existing methods on 170 synthetic and natural images.
• Achieved state-of-the-art results by combining region-based, feature-based, and frequency-based approaches.

Conclusions:

• The proposed algorithm effectively detects complex, affinely skewed rotation symmetries.
• The frieze-expansion method offers a robust framework for advanced symmetry analysis.
• This work advances the field of image symmetry detection with a comprehensive approach.