Fast 3D Object Segmentation in X-ray Tomography
Robert M. Evans, Sheridan C. Mayo · 2009
Abstract: Segmenting objects or internal structures from volumetric images has found many applications in X-ray computer tomography. Owing to the data-intensive nature of these applications, increasing computational efficiency becomes a primary concern for the design and development of 3D object segmentation methods. We propose an efficient deformable model with multiple templates to segment 3D vessel objects in X-ray tomographic images of wood. In our approach, an automatic vessel segmentation method is applied to the first image of the 3D sequence to obtain the initial template for object contours. A deformable energy minimization concept is applied, involving a balance between internal forces from spline contour stretching and bending and external forces from salient image features. The template then interacts with the adjacent frame and deforms iteratively toward the local target, which can be preserved as the input for the next frame in the sequence. This process is repeated for consecutive frames until the object leaves the field of view or a new object emerges. The deformable model is fast and accurate in tracking small object variation in a continuous space. However in our application, vessel diameters and orientations in most hardwoods can vary considerably along the longitudinal axis i.e. scanning direction. This may result in large variations in vessel shape and size from frame to frame, making it difficult to devise appropriate continuity constraints for