Classification of partial 2-d shapes using fourier descriptors.

Area of Science:

• Computer Vision
• Image Analysis
• Pattern Recognition

Background:

• Shape classification is crucial in various fields.
• Handling incomplete or partial shape data presents significant challenges.
• Existing methods struggle with arbitrary rotations, scaling, and missing data.

Purpose of the Study:

• To develop a robust method for 2-D partial shape classification.
• To estimate Fourier descriptors of complete shapes from incomplete observations.
• To improve classification accuracy despite missing boundary data.

Main Methods:

• Utilizing Fourier descriptors for shape representation.
• Formulating the problem as estimating unknown Fourier descriptors.
• Minimizing a cost function including least squares fit and shape regularity (perimeter^2/area).

Main Results:

• Accurate estimation of Fourier descriptors for complete shapes from partial data.
• Successful classification experiments with both synthetic and real boundary data.
• Achieved reasonable classification accuracies even with 20-30% data missing.

Conclusions:

• The proposed method effectively classifies 2-D partial shapes.
• Robustness to arbitrary rotation, scaling, and missing data is demonstrated.
• The technique offers a viable solution for shape analysis with incomplete information.