Finding methods for detecting objects in computer tomography images has been an active area of research in the medical and industrial imaging communities. While the raw image can be readily displayed as 2-D slices, 3-D analysis and visualization require explicitly defined object boundaries when creating 3-D models. A basic task in 3-D image processing is the segmentation of an image that classifies voxels/pixels into objects or groups. It is very computation intensive for processing because of the huge volume of data. The objective of this research is to find an efficient way to identify, isolate and enumerate 3-D objects in a given data set consisting of tomographic cross-sections of a device under test. In this research, an approach to 3-D image segmentation and rendering of CT data has been developed. Objects are first segmented from the background and then segmented between each other before 3-D rendering. During the first step of segmentation, current techniques of thresholding and image morphology provide a fast way to accomplish the work. During the second step of segmentation, a new method based on the watershed transform has been developed to deal with objects with deep connections. The new method takes advantage of the similarity between consecutive cross section images. The projections of the objects in the first image are taken as catchment basins for the second image. Only the different pixels in the second image are processed during segmentation. This not only saves time to find catchment basins, but also splits objects with deep connections that cannot be simply implemented by the watershed transform. A unique label has been issued to each object after segmentation. Objects can be distinguished well from each 2-D slice by their labels. This is a good preparation for 3-D rendering and quantitative analysis of each object. In this thesis, a novel 3-D rendering has been developed by surface rendering approach. A new and easier rendering model has been invented under the assumptions that light comes from the same side as the viewer, both of which are situated at infinity. It works fast because only surface pixels are being processed and interior pixels are left unprocessed. The surface intensity of the objects is attenuated by coefficients according to their distance from the viewer. The objects finally are shown from top and side views. Volume rendering was accomplished by sample images as well. In this research, the new method works several times faster than previous methods. After successful segmentation and rendering, the volume of each object can be easily calculated and the objects are recognizable in 3-D visualization. Keywords: 3-D Image Segmentation, 3-D Image Rendering, Watershed Transform, Surface Rendering, Thresholding, Morphological Transform.
Library of Congress Subject Headings
Image processing--Digital techniques; Three-dimensional imaging; Image reconstruction; Tomography
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Wang, Hui, "3-D image segmentation and rendering" (2003). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus