Abstract: The existence of structural texture can interfere with the defects (such as scratches) detection for the surface texture images defects detection of machined surface. The method of eliminating background texture in frequency domain was studied for the defects detection. Firstly, based on the image restoration technology of Fourier transform, the structural texture in spatial domain image corresponded to the high energy frequency component in Fourier domain, which was removed by the least square method linear fitting, and these energies were set to zero finally transformed into the spatial domain image. In the restored image, the structural texture region in original image became an approximate uniform gray level, in which the defective region was preserved. And then the statistical process control was used to set threshold to separate the defects from the restored image. Finally, experiments on a series of structural texture images show that the proposed method is feasible and effective.