LAUSR

Abstract This paper presents a machine vison method based on mathematical morphology for defect detection in machined surfaces that contain circular tool-marks. The traditional morphology with rectangular-shaped structuring elements (SE) has been applied successfully for defect detection in the surfaces with linearly structured patterns. In order to apply the traditional morphology for circularly-textured surfaces of a circular machine part, the polar-coordinate conversion is required. It creates noise and artifacts in the polar-transformed image due to pixel value interpolation. The morphological operations with arc-shaped SEs are thus proposed in this study, which make the morphology directly applicable to the original input image without polar transformation. The table-look-up technique is used for the retrieval of the x-y coordinates of all members in an arc-shaped SE of arbitrary size at arbitrary location in the input image. It makes the computation of morphological operations with arc-shaped SEs as fast as that with rectangular-shaped SEs. The proposed morphological operations can efficiently intensify local defects and remove the tool-mark background in the circular machined surface. The experimental results show that the proposed method can achieve high detection accuracy for various small defects, including scratch, bump and edge burst. It is also computationally efficient since it only requires 0.2 s to process a 612 × 612 test image on a typical personal computer.