For detection of the end face defects of small cylinder core with complex image background, the current detection methods based on threshold, morphological and edge Canny are inaccurate due to the interference of the gray information caused by the unevenness of the surface coating. We present a defect detection algorithm based on machine vision for optical triangulation. Firstly, two images with certain defect information are obtained based on the active light visual inspection method. Then, the defect contours are sketched out according to the template matching, morphological analysis and frequency statistics. Finally, we use Delaunay triangulation technique to realize the accurate sketch of the defect area. The experimental results show that this method has high accuracy and robustness for defect detection hidden in complex background, and the repeatability can reach sub-pixel level, with good anti-jamming, strong practicability and high efficiency.