Deep learning phase shift interference fringe pattern blind denoising method for interferometry

Abstract

The invention discloses a deep learning phase shift interference fringe pattern blind denoising method for interferometry, and belongs to the field of image processing. The method comprises the following steps: S1, generating a series of sample arbitrary phase shift noise interference fringe patterns and corresponding sample arbitrary phase shift noise-free interference fringe patterns; s2, aftercutting, constructing a sample data set; s3, designing a deep learning phase shift interference fringe pattern blind denoising convolutional neural network framework; s4, obtaining a trained deep learning phase shift interference fringe pattern blind denoising convolutional neural network model; and S5, obtaining any phase shift noise-free interference fringe pattern of the sample. The method inputs a sample arbitrary phase shift noise interference fringe pattern into a trained deep learning phase shift interference fringe pattern blind denoising convolutional neural network model, and outputsa sample arbitrary phase shift noise-free interference fringe pattern with high efficiency and high quality; the method provided by the invention is good in blind denoising quality and high in speed,improves the noise removal efficiency, and has wide practical value and application prospect.

Description

technical field [0001] The invention belongs to the field of image processing, and in particular relates to a blind denoising method for deep learning phase-shifting interference fringe images used in interferometry. Background technique [0002] At present, interferometry has become a new benchmark for quantitative phase imaging and interference microscopy due to its non-contact, non-destructive, high-precision and quantitative original imaging advantages, and it is also used in 3D imaging of microscopic topography and surface defect detection. The field has been extensively applied research. However, in phase-shift interferometry, due to the dark field, high temperature of the image sensor, and unstable environmental disturbances, the phase-shift interference fringe pattern recorded by the CCD must contain noise. Generally speaking, the phase-shift noise interference fringe pattern will affect the accuracy of phase recovery, resulting in a decrease in the quality of phase...

AI Technical Summary

Problems solved by technology

For example, a two-dimensional windowed Fourier filtering method is proposed to reduce noise in the fringe pattern and the exponential phase field, but this method may cause boundary artifacts in the denoised fringe pattern; a method using image decomposition is proposed Interference fringe image denoising method, but this method requires prior knowledge of dividing the interference fringe image into different spaces; a noise reduction method based on dimensionality reduction is proposed, which mainly uses the singular value decomposition method to suppress the interference fringe image noise, but this method requires additional operations to perform rotation and zero-filling processes; a block-matching 3D filtering method is proposed, which uses a non-local adaptive non-parametric filtering method for image denoising by grouping and collaborative filtering, but the The method first needs to estimate the noise level, and this method is mainly for general images with a specific noise level

Although all the above methods can achieve satisfactory denoising effect, how to effectively remove the noise in the interference fringe pattern is still a key issue because the noise in the interference fringe pattern is random and unknown

In the summary of the existing interference fringe image denoising methods, it is found that the existing methods have the following disadvantages: denoising image quality is low, slow, requires prior knowledge, needs to perform additional operations, and targets specific noise levels, etc.

Images