Abstract: In order to solve the problem that it was hard to extract the outline information from the magneto-optical image of welding defects, an image denoising and contour detection method was proposed. The crack defect samples were obtained by laser spot welding, and the magnetic flux leakage field of defects was simulated by finite element method. The gray continuity, concentration, and noise characteristic of magneto-optical images were compared and analyzed. The fast non-local mean filtering algorithm was used to remove noise, and compared with the traditional filters. According to the distinction in gradient value between flaw information and background information in magneto-optical images, an operator was used to carry out Otsu threshold segmentation and the edge detection based on the first order gradient graph of the image. The results show that the standard deviation and image entropy of the magneto-optical image processed by this method reaches 30.0465 and 6.0395 respectively, and the image aggregation degree is better, which is closer to the simulation magnetic field curve, and the defect contour information could be better extracted. This result is helpful for defect recognition and target detection.