Weld defect detection by X-ray images method based on Fourier fitting surface

To solve such problems as the strong noise, low contrast and complex background of X-ray image in the weld defect detection, a method of the noise reduction, weld edge segmentation and defects extraction was proposed. The fast discrete curvelet transform and cycle shift were applied to reduce noise of the weld image, and the Otsu method was utilized to extract the weld region by the column gray curves of the image. Cubic Fourier curve was used to fit the column gray curves after preprocessing of weld image, and the adaptive threshold surface was constructed by extending fitting curves to 3D space. Finally, the background and the defect area were segmented accurately with the gray differences of 3D gray image between the original image and the reconstructed surface. Experiment results show that the method can extract weld defects accurately. Compared with traditional defect detection algorithm, it has the lower undetected rate and fewer misinterpretations, the accuracy rate could reach 95%.