A refined denoising method for noisy phase-shifting interference fringe patterns

• Published in: Optical and quantum electronics Vol. 53; no. 8
• Main Author: Omar, E. Z.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.08.2021; Springer Nature B.V
• Subjects: Accuracy; Birefringence; Characterization and Evaluation of Materials; Classification; Computer Communication Networks; Datasets; Diffraction patterns; Efficiency; Electrical Engineering; Interference fringes; Lasers; Machine learning; Noise; Noise reduction; Nylon 6; Optical Devices; Optics; Phase distribution; Photonics; Physics; Physics and Astronomy; Random noise; Support vector machines; Phase-shifting interferometry (PSI) techniques; Denoising method; Birefringence; Speckle noise; iPP fibres; Deep learning techniques
• ISSN: 0306-8919; 1572-817X
• DOI: 10.1007/s11082-021-03106-4

Denoising process is indispensable procedure to get accurate wrapped and unwrapped phase maps from noisy phase-shifting interference fringe patterns. The efficiency of this process has high correlation with the classification accuracy of the noise type. Usually, the classification process for the noise type in the fringe patterns is performed via the experience which may increase the error level. In this paper, a denoising method based on a refined pre-trained deep learning networks is proposed. This method can perform an automatic denoising for the noisy fringe patterns according to its type. So that, the pre-classification process for the noise type is the core of this method. A dataset containing 902 numerical stimulated interference fringe patterns is established. This dataset includes four classes; no noise, salt and pepper noise, Gaussian noise and speckle noise. To perform the automatic classification process, a pre-trained Alex CNN is fine-tuned. This network achieved 97.5% validation accuracy and 98% testing accuracy. To improve the efficiency, an incorporation between the AlexNet and the support vector machine classifier is proposed. After the classification process, each noisy interference fringe pattern is treated according to its class. The proposed method is applied on realistic noisy interference patterns for isotactic polypropylene (iPP) and nylon 6 fibres captured using the phase-shifting interference microscope. The phase distribution values and the 3D birefringence are calculated for iPP and nylon 6 fibres. Our experimental results show that our proposed method achieves high efficiency in the denoising process.