In this study, to detect and evaluate welding defects effectively, magnetic flux leakage characteristics of magneto-optic imaging under alternating magnetic field excitation are proposed for application in the contour reconstruction of welding defects. A magnetic flux leakage reconstruction model was also established to study the two-dimensional contour characteristics of welding defects. First, based on the formation mechanism of a leakage magnetic field under an alternating magnetic field, the relationship between the two leakage magnetic field component signals (By and Bz) and the defect contour was discussed. Second, a generalized regression neural network was trained using numerical simulation data, to determine the model and to show that the leakage magnetic field signal can achieve defect contour reconstruction. Finally, the data derived from magneto optic imaging magnetic leakage characteristics were applied to the training of the model, to determine the feasibility of reconstruction. Experimental results show that the image data of the magnetic flux leakage characteristics are consistent with the contour reconstruction rule obtained through simulation, and a two-dimensional contour reconstruction of welding defects can be realized. Within a specific range, the greater (no less than 0.45 mm) the depth of the defect, the better the reconstruction effect.