This paper introduces the components of the hardware, software, working principle and the defect detection capacity of computer vision system for x-ray real time radiography weld defect detection and evaluation. Meanwhile, the image processing system is studied in which an image transformation algorithm is presented for defect extraction. Moreover, a sectional curve fitting algorithm and an algorithm based on probability are proposed to locate the defects. At the end of the paper, a pattern recognition method is introduced for the automatic identification of weld defects.