摘要

A new approach is proposed for automatically detecting crack defects with dark colors and low contrasts in magnetic tile images using the fast discrete curvelet transform (FDCT) and texture analysis. In this methodology the original images were first decomposed and reconstructed based on the FDCT. Then the thresholds of decomposition coefficients were calculated by texture feature measurements. With these thresholds the surface textures in the images can be eliminated. Finally by extracting contours from the reconstructed images, the expected images without textures but with crack defects contours were obtained. Experimental results show that the proposed method could eliminate the contours of the textures, and extract from the image cracks longer than 0.8 mm.