Abstract: Aiming at the problem of irregular shape and low contrast of the scratches in the image of mobile phone screen, a method based on machine vision was proposed to detect the surface scratches of mobile phone screen. Firstly, the PatMax algorithm and affine transformation were adopted to preprocess the screen images of mobile phone. Then the shear transformation was used to decompose the image into two parts: low frequency and high frequency, the element shapes in four directions of 0°, 45°, 90° and 135° were constructed to perform gray-scale closing operation on the low frequency part, and the N × M median filter denoising operation was performed on the high frequency part, the enhanced image was generated by the inverse shear transformation. Finally, the improved Otsu double threshold method was used to extract the target. 450 pieces of mobile phone screen images were randomly selected for experiments, and the highest detection rate is 98.7%. The results show that this method can effectively enhance the detail information of the images, which greatly guarantees the integrity of the scratch defects.