A novel automated approach for noise detection in interference fringes pattern images using feature learning

This study presents an automated system to effectively identify different noise types and levels of interference fringe pattern images. The key idea involves feature extraction from noise samples using extra filters and multilayer neural network. Besides median, wiener and homomorphic filter, NL-means filter is used to separate noise samples based on non-local self-similarity of fringe patterns. Statistical methods like kurtosis and skewness are extracted and used for neural network learning. The system is capable of accurately classifying the type and level of noise of fringe patterns and specific filter can be applied. The experiment result shows that the accuracy of high noise level is still over 82%. We introduce non-local fractional-order diffusion equation filtering method for high level Gaussian noise corrupted electronic speckle pattern interferometry fringes denoising. The proposed method is based on partial differential equation (PDE) and non-local methods. The first term of the energy functional is nonlinear P-M function which can remove noise meanwhile preserve edges. The second term is fractional order total variation energy functional, it can use the selfsimilarity of fringes pattern and improve the quality of the denoised image.