Abstract: A new improved segmentation method based on active contour models was proposed to solve the problems, like uneven illumination, low-rise contrast, blurry defect edge and other drawbacks in the X-ray welding image. An average shift histogram was introduced into the traditional active contour models to estimate the nonparametric probability, to make the model be driven by both gradient forces and statistical pressure, so as to solve the problems in the initialization and accurate segmentation about welding joints defects. The experiment results show that, the proposed method can extract welding joints defects from the X-ray welding joints image accurately, and improve segmentation accuracy.