Computational de-noising based on deep learning for phase data in digital holographic interferometry

Abstract

This paper presents a deep-learning-based algorithm dedicated to the processing of speckle noise in phase measurements in digital holographic interferometry. The deep learning architecture is trained with phase fringe patterns including faithful speckle noise, having non-Gaussian statistics and non-stationary property, and exhibiting spatial correlation length. The performances of the speckle de-noiser are estimated with metrics, and the proposed approach exhibits state-of-the-art results. In order to train the network to de-noise phase fringe patterns, a database is constituted with a set of noise-free and speckled phase data. The algorithm is applied to de-noising experimental data from wide-field digital holographic vibrometry. Comparison with the state-of-the-art algorithm confirms the achieved performance.

Lock-in vibration retrieval based on high-speed full-field coherent imaging

“…From that the noisy Doppler phases are processed to get the amplitude and phase of the vibration. Three de-noising algorithms are considered: first, the median filter 57 that is considered for comparison purpose 58 , then the windowed Fourier transform which is a filter operating by applying threshold in the Fourier domain 59 (noted WFT2F), last a filter operating by Deep Learning approach 60 (noted DeepL). The way to process the holographic data with the VibMap is considered with, first the direct application to noise-free simulated data, second, the direct application to the noisy data, third application to noisy data followed by the de-noising algorithm, and last the de-noising algorithm followed by the application of VibMap .…”

“…At the heart of this new tool is the convolutional neural network (CNN) which integrates several fundamental advances of last decades: wavelet and multiresolution analysis, shrinking and thresholding algorithms, sparse representations, block matching and dictionary learning 76 , 77 . In recent years, several applications of deep learning in optics have emerged, the deep learning approach is applied to noise reduction and to enhance the quality of the reconstructed tomographic image quality 60 , 78 , 79 . In previous works, the problem of speckle decorrelation noise was approached with a deep learning based solution 60 exhibiting performances at the state-of-the art for speckled phase data de-noising.…”

Lock-in Vibration Retrieval Based on High-speed Full-field Coherent Imaging

“…From that the noisy Doppler phases are processed to get the amplitude and phase of the vibration. Three de-noising algorithms are considered: first, the 7 × 7 median filter 54 that is considered for comparison purpose 55 , then the windowed Fourier transform which is a filter operating by applying threshold in the Fourier domain 56 (noted WFT2F), last a filter operating by Deep Learning approach 57 (noted DeepL). The way to process the holographic data with the VibMap is considered with, first the direct application to noise-free simulated data, second, the direct application to the noisy data, third application to noisy data followed by the de-noising algorithm, and last the de-noising algorithm followed by the application of VibMap.…”

“…At the heart of this new tool is the convolutional neural network (CNN) which integrates several fundamental advances of last decades: wavelet and multiresolution analysis, shrinking and thresholding algorithms, sparse representations, block matching and dictionary learning 69,70 . In recent years, several applications of deep learning in optics have emerged, the deep learning approach is applied to noise reduction and to enhance the quality of the reconstructed tomographic image quality 57,71,72 . In previous works, the problem of speckle decorrelation noise was approached with a deep learning based solution 57 exhibiting performances at the state-of-the art for speckled phase data de-noising.…”

“…In recent years, several applications of deep learning in optics have emerged, the deep learning approach is applied to noise reduction and to enhance the quality of the reconstructed tomographic image quality 57,71,72 . In previous works, the problem of speckle decorrelation noise was approached with a deep learning based solution 57 exhibiting performances at the state-of-the art for speckled phase data de-noising. The network based on residual learning 73 has been adapted to be trained with a set of fringe patterns including realistic speckle noise conditions, similarly as what is usually encountered in coherent phase imaging.…”