Abstract: It is difficult to detect the surface defects of the motor magnetic tile, because the image contrast is low, the noise is serious, and there are bright stripes. Aimed at this problem, an anisotropic diffusion model with bright region suppression function is proposed. A bright region feature description operator was designed to distinguish defects and interference, and a new diffusion coefficient function was constructed. The experimental results show that compared with similar algorithms, the model has obvious advantages in background smoothing, edge detection, image segmentation and defect recognition. The detection precision of blowhole and crack defects can reach 95% and 89% respectively. This research effectively enhances the low contrast image of the magnetic tile, improves the defect detection precision, and has high engineering application potential.