In order to detect a weld defect using machine vision techniques, it is necessary to ascertain occurrence of noise and estimate the noise level in a weld X-radiography image. In general, noise, particularly, Gaussian or normal noise, appears in an image during the image acquisition process. While implementing image-processing techniques, the presence of Gaussian noise in an X-radiography image degrades the desired result. Therefore, we should ensure the presence of noise in the test image. To solve this problem, this study proposes a new scheme in which a single image can be taken instead of two, i.e., an original noise-free and a noisy image. We employed the anisotropic diffusion technique proposed by Perona-Malik for image smoothing and noise level estimation using standard deviation. The standard deviation between the input and smoothened image confirms the presence of noise in the test image. The proposed scheme delivers accurate results, irrespective of the presence or absence of noise, and estimates noise level.

noise detection, noise identification, noise-level estimation, anisotropic diffusion, weld defect, X-radiography image.