The FLANN（ Fast library for approximate nearest neighbors） algorithm is used to match the detected SURF（ Speeded up robust features） descriptors. The obtained key points are iteratively sorted,and the RANSAC（ Random sample consensus） algorithm is used to extract the matching points for two times. The gray difference algorithm is used to extract the defective region after the matrix transformation registration,using the Otsu threshold method to segment the image and the characteristics of defect area domain are analyzed,which uses the Otsu threshold method to segment the image. The results show that the method is superior to the traditional detection algorithm in detecting speed and range. So it has positive research significance and value in on-line inspection of small module plastic gear industry.