Segmentation of industrial CT image using local robust statistics and 3D C-V model

The segmentation of 3D images has important applications in the non-destructive examination of industrial computed tomography (CT) and has been widely studied. When a C-V model is used to segment the image, closed and accurate object contours can be obtained. The C-V model is suitable for image segmentation and extended easily to a 3D application. However, industrial CT images may have artifacts or noise, which may make active contours stop at undesired boundaries. In order to overcome these difficulties, the 3D C-V model is improved by the robust statistics method. In this improved model, the intensity of the image in the 3D C-V model is replaced with robust statistics, which is the weighted combination of the inter-quartile range, mean absolute deviation and intensity median in the local region of the image. Inter-quartile range and mean absolute deviation in the local region are introduced to sharpen object boundaries, and the intensity median in the local region reduces image noise. The comparison experiments demonstrate the position accuracy of the contours and robustness to noise of the improved model.