Enhancing polarization 3D facial imaging: overcoming azimuth ambiguity without extra depth devices

Area of Science:

• Computer Vision
• Optical Imaging
• 3D Reconstruction

Background:

• Polarization 3D imaging offers biosafety, efficiency, and simplicity for 3D facial reconstruction.
• A key limitation is the multi-valued problem of the normal vector's azimuth angle.
• Existing solutions often require additional depth-capturing devices, limiting applicability.

Purpose of the Study:

• To present a passive 3D polarization facial imaging method that overcomes azimuth ambiguity without additional depth devices.
• To leverage facial feature information for accurate normal vector determination.
• To achieve precise 3D facial reconstruction with preserved details.

Main Methods:

• Developed a passive 3D polarization facial imaging technique.
• Utilized statistical analysis of azimuth angle distributions and facial feature detection.
• Implemented an azimuth angle correction method based on facial feature point locations.

Main Results:

• Accurate assignment of polarized normal azimuth angles to quadrants using feature detection.
• Achieved precise unique normal vector determination and accurate 3D facial reconstruction.
• Validated correction for over 75% of pixels in simulated data without extra depth techniques.

Conclusions:

• The proposed method enables polarization 3D facial imaging under natural conditions without extra depth devices.
• Facial feature-based azimuth ambiguity resolution is effective for 3D reconstruction.
• The technique preserves crucial edge details and texture information in the reconstructed 3D faces.