Reflection symmetry detection using locally affine invariant edge correspondence

Area of Science:

• Computer Vision
• Image Processing
• Computational Geometry

Background:

• Reflection symmetry detection is crucial in computer vision.
• Current methods often rely on intensity-based features (e.g., SIFT), which are sensitive to illumination variations and struggle with textureless objects.
• There is a need for more robust symmetry detection algorithms.

Purpose of the Study:

• To propose a novel reflection symmetry detection algorithm.
• To overcome the limitations of intensity-based feature matching.
• To achieve robust symmetry detection under varying illumination and object distortions.

Main Methods:

• The proposed method utilizes locally affine invariant edge-based features.
• Affine invariance is achieved through efficient linear algebra computations.
• A complete pipeline involves edge detection, feature description and matching, and a voting process for symmetry axis identification.

Main Results:

• The novel approach demonstrates superior performance compared to existing intensity-based methods.
• The algorithm is insensitive to illumination variations and effective for textureless objects.
• It shows robustness against object distortions, including perspective projection.

Conclusions:

• The developed edge-based feature correspondence method offers a more robust and versatile solution for reflection symmetry detection.
• This technique enhances the applicability of symmetry detection in challenging real-world scenarios.
• The findings validate the proposed method's effectiveness through experiments on synthetic and real-world images.