Abstract: In order to achieve the location and visualization of defects in T-shaped weldments, defect segmentation of X-ray detection image must be done accurately. Based on the analysis of the grey distribution curve of micro-focus X-ray detection images of T-shaped weldment in horizontal and vertical directions, the column pixels of image were processed by single column sliding template step by step, defects and noise were smoothed and filtered, the background simulation of the detected images was achieved successfully. Combining with the method of threshold segmentation, the segmentation of weld defects were completed accurately, and the results were compared with those of the two background simulation methods of morphology and mean filter. The experimental results show that the weld defects in the detection images are segmented effectively by the proposed algorithm, and it can avoid the problem such as missed detection, over segmentation and so on caused by other simulation methods. At the same time, it can provide good basis for subsequent defects thinning and automated extraction of spatial position data.