Unifying temporal phase unwrapping framework using deep learning.

Area of Science:

• Optics and Photonics
• Computer Vision
• Metrology

Background:

• Temporal phase unwrapping (TPU) is crucial for 3D surface reconstruction in fringe projection profilometry, especially for discontinuous or isolated objects.
• Existing TPU algorithms (multi-frequency, multi-wavelength, number-theoretic) require multiple fringe patterns, increasing complexity and reducing speed.
• Image noise significantly degrades the accuracy and efficiency of traditional TPU methods, necessitating numerous auxiliary patterns.

Purpose of the Study:

• To introduce a generalized deep learning framework for temporal phase unwrapping (TPU).
• To demonstrate the framework's ability to enhance reliability and mitigate noise across different TPU algorithm groups.
• To improve phase retrieval efficiency without increasing the number of required fringe patterns.

Main Methods:

• Development of a generalized deep learning framework applicable to various TPU algorithm categories.
• Implementation of the framework to process fringe patterns obtained from fringe projection profilometry.
• Experimental validation of the deep learning approach against traditional TPU methods.

Main Results:

• The deep learning framework effectively mitigates the impact of image noise on phase unwrapping.
• Phase unwrapping reliability is significantly enhanced across different TPU approaches.
• The method achieves improved performance without requiring an increased number of auxiliary fringe patterns.

Conclusions:

• Deep learning provides a powerful and generalized solution for temporal phase unwrapping (TPU).
• The proposed framework enhances measurement accuracy and speed in fringe projection profilometry by reducing noise sensitivity.
• This approach holds significant potential for advancing reliable phase retrieval techniques.