CT Multi-Layer Cylindrical Object Delamination Analysis Using Randomized Hough Transformation

In our previous paper[1], we described a CT layer segmentation approach capable of automatically segmenting internal planar layers from volumetric data. In this paper, we extend our work to the segmentation of curved surfaces, e.g. cylindrical surfaces, by using a novel reduced dimension randomized Hough transformation. First subvoxel features (edges with polarities indicating rising or falling edges) are detected from each slice with an energy balanced transition detection algorithm. Then a randomized Hough transformation is applied to calculate five parameters of an ellipse in 3D space. Instead of constructing a five dimensional Hough Space, we propose a unique solution by using five 1D Hough spaces and one 1D verification space to locate Hough peaks. The rotation axis of a cylindrical object can then be determined by fitting a sequence of found ellipses. Finally, the same ellipse fitting method is applied to cut planes along the rotation axis to determine layer positions. All detected ellipses are classified into different layers based on edge polarities and sizes. We can unwrap cylindrical layers into planar surfaces for delamination analysis. The proposed method has been applied to a 3 layer cylindrical object. All three layers have been detected, unwrapped and analyzed for potential delamination detection in 40 seconds (512 cube using a notebook).