DeepOrientation: convolutional neural network for fringe pattern orientation map estimation.

Area of Science:

• Optical Metrology
• Computational Imaging
• Biomedical Optics

Background:

• Fringe pattern analysis is vital for full-field optical metrology at macro and micro scales.
• Accurate local fringe orientation maps are crucial for enhancing measurement processes like denoising and phase demodulation.
• Existing methods for orientation estimation can be limited in robustness and automation.

Purpose of the Study:

• To introduce DeepOrientation, a novel deep learning-based numerical solution for accurate and robust local fringe orientation map estimation.
• To demonstrate the effectiveness of DeepOrientation compared to classical methods.
• To showcase its application in guiding phase demodulation for label-free microscopy.

Main Methods:

• Development of a convolutional neural network (CNN) model for fringe orientation estimation.
• Numerical simulations to validate the proposed DeepOrientation method.
• Experimental validation using fringe pattern analysis and quantitative phase imaging of living cells.

Main Results:

• DeepOrientation significantly outperforms the combined plane fitting/gradient method in numerical simulations and experimental tests.
• The method provides accurate fringe orientation maps, facilitating subsequent image processing tasks.
• Successful application in guiding Hilbert spiral transform for phase demodulation in label-free microscopy.

Conclusions:

• DeepOrientation offers a robust, accurate, and automated solution for local fringe orientation estimation in optical metrology.
• The method is a valuable tool for various fringe pattern analysis applications, particularly in label-free quantitative phase microscopy.
• DeepOrientation enhances the capabilities of advanced imaging techniques for biological samples.