Image Analysis for Specific Region Segmentation of Steel-tube Radiographic Images

This paper describes an image analysis for region segmentation from gray-scale images of steel-tube radiographic detection without the usual step of binarization. A multi-stage approach is presented for boundary detected images, along with industrial application. Global contrast analysis is used in first stage to give a preliminary result, boundary and range. The second stage is a lineal segmentation in which both pixel location and intensity similarity are taken into account. Next stage then refines the threshold value based on local differentiation. By applying the modified fuzzy c-means algorithm, the spatial partitioning of a lineal pixels' set into sub-regions becomes context-oriented and fully iterative, unlike conventional spacial process. Analyzing results indicate that the steeltube radiographic images can be segmented into specific regions, even though noise, illumination and presence of shadows.