Single Fringe Phase Retrieval for Translucent Object Measurements Using a Deep Convolutional Generative Adversarial...

Area of Science:

• Optics and Photonics
• Computer Vision
• Metrology

Background:

• Fringe projection profilometry (FPP) is vital for 3D reconstruction but struggles with translucent objects due to phase shift and reduced fringe modulation.
• Surface scattering effects in translucent materials degrade measurement accuracy in FPP.

Purpose of the Study:

• To introduce GAN-PhaseNet, a novel single-frame phase retrieval method to enhance 3D measurement accuracy for translucent objects.
• To mitigate the impact of surface scattering on wrapped phase measurements in FPP.

Main Methods:

• Developed GAN-PhaseNet, a generative adversarial network incorporating U-net++, Resnet101 backbone, and a multilevel attention module.
• Employed a single-frame approach for phase retrieval to simplify the measurement process.

Main Results:

• GAN-PhaseNet achieved superior phase retrieval accuracy, matching conventional methods for non-scattering and slightly scattering objects.
• The method demonstrated the lowest errors for severely scattering objects compared to CDLP, Unet-Phase, and DCFPP.
• GAN-PhaseNet exhibited excellent robustness and generalization across varying noise levels and fringe frequencies.

Conclusions:

• GAN-PhaseNet effectively improves 3D reconstruction accuracy for translucent objects in fringe projection profilometry.
• The proposed method offers a robust and accurate solution for computational imaging in FPP, particularly for challenging translucent materials.