The work aims to accurately detect the defects of printed words on the medicinal glass bottles. The currently popular SIFT feature point extraction operator was adopted. Then, the Euclidean distance was used for the initial matching, and the RANSAC was used for the exact matching. In the transmission process, no matter how the medicinal bottle changed, the defects of the printed words could be detected. Thus, the unqualified medicinal bottles could be successfully removed. The experiment results show that this method can accurately extract the image feature points and match the image feature point pairs.