Three-Dimensional Face Recognition Using Solid Harmonic Wavelet Scattering and Homotopy Dictionary Learning.

Area of Science:

• Computer Vision
• Pattern Recognition
• 3D Data Analysis

Background:

• 3D data representation (point clouds/meshes) poses challenges due to noise and non-uniformity.
• Existing 3D face recognition methods struggle with environmental variations like posture and illumination.

Purpose of the Study:

• To propose a robust geometric deep learning framework for 3D face recognition.
• To address the limitations of traditional methods in handling raw, non-uniform 3D scan data.
• To achieve high accuracy in 3D face recognition despite variations.

Main Methods:

• Utilized a geometric deep learning framework processing raw spatial coordinates.
• Modeled 3D face scans as stochastic processes to handle non-uniformity.
• Applied a windowed solid harmonic scattering transform to extract invariant coefficients.
• Employed a sparse learning network for classification and reducing intraclass variability.

Main Results:

• Achieved a rank-1 recognition rate (RR1) of 99.84% on the Face Recognition Grand Challenge (FRGC) v2.0 dataset.
• Achieved a rank-1 recognition rate (RR1) of 99.90% on the Bosphorus dataset.
• Demonstrated superior performance compared to existing methods, even with fragmentary or deformed samples.

Conclusions:

• The proposed geometric deep learning framework offers a robust solution for 3D face recognition.
• The method effectively handles variations in 3D face scans, achieving state-of-the-art performance.
• This approach advances the field of 3D pattern recognition and computer vision.