Pseudodistance measures for recognition of curved objects

Area of Science:

• Computer Vision
• Geometric Modeling
• Pattern Recognition

Background:

• Describing and identifying complex curved objects remains a challenge in computer vision and pattern recognition.
• Existing methods may struggle with the intricate details and variations found in curved object shapes.

Purpose of the Study:

• To develop a novel method for the accurate description and identification of curved objects.
• To represent object shape as a surface derived from horizontal section boundaries.
• To decompose complex shapes into a set of fundamental geometric features.

Main Methods:

• Representing object shape using a Fourier series expansion of a representative function.
• Decomposing object features into elongatedness, horizontal strain, section shape, torsion, and displacement.
• Defining object differences as pseudodistances between these decomposed features.
• Theoretically analyzing properties of the pseudodistance family, including completeness.

Main Results:

• The proposed method successfully describes and identifies curved objects based on their decomposed features.
• Experimental validation using wooden block models and a doll demonstrated the method's effectiveness.
• The theoretical analysis confirmed desirable properties of the defined pseudodistance metric.

Conclusions:

• The Fourier-based decomposition into five features provides a robust way to describe curved object shapes.
• The pseudodistance metric enables effective pattern recognition and identification of curved objects.
• This method offers a promising approach for applications requiring precise shape analysis and object recognition.