The work aims to propose a bottle cap scratch detection scheme based on machine vision technology so as to achieve fast and accurate detection of fine scratches in a low contrast background since the current domestic bottle cap scratch detection methods do not have high precision under the complicated background conditions, the gray value of the bottle scratch image is changed sharply, and many influencing factors cannot be accurately positioned. The standard image was preprocessed to create the template, and the sample image was filtered to reduce noise. The template was matched according to shape, the region of interest (ROI) was extracted, Gaussian difference filtering was used to enhance the contrast of the scratch, the two-dimensional Otsu threshold segmentation was performed, morphological processing was conducted, and scratches were extracted according to features. By obtaining the image of the surface of 300 caps, the methods of aberration, Dajin and manual detection were used to compare the scratches. The experimental results showed that the proposed algorithm could extract the cap scratches quickly, accurately and efficiently. The average time for detecting a picture was 113 ms, and the detection accuracy was 98.3%. Compared with manual detection, differential image method and Dajin method, the scheme has higher detection precision, faster speed and better robustness, and meets industrial production requirements.