In order to analyze the three-dimensional information such as the position, shape and direction of the ceramic bowl, the reconstruction algorithm of a local point cloud based on image sequences is proposed. Firstly, a calibrated binocular camera is used to collect many images of surface defect from different angles. Then the image feature point detection and matching algorithm of FAST+SURF+FLANN is used to get high precision matching point pairs. Finally, the local three-dimensional reconstruction of two-dimensional surface defects is realized through the structure from motion algorithm combined with patch-based multi-view stereo algorithm. However, the direction and location of the defect cannot be perfectly described through the local three-dimensional reconstruction. So the proposed algorithm of manually added feature points is adopted to realize the global reconstruction of the surface of the ceramic bowl. The results show that the defect reconstruction results are clear, and the defect position, direction and shape information is complete.