Piecewise-Linear Interpolation between Polygonal Slices (original) (raw)

1996, Computer Vision and Image Understanding

netic resonance imaging) apparata. These cross sections, In this paper we present a new technique for piecewise-linear hereafter called slices, are the basis for interpolating the surface reconstruction from a series of parallel polygonal cross boundary surface of the organ. The interpolated object sections. This is an important problem in medical imaging, can then be displayed in graphics applications, or (more surface reconstruction from topographic data, and other applirecently) even manufactured by an NC (numerically concations. We reduce the problem, as in most previous works, to trolled) or an RP (rapid prototyping) machine. Another a series of problems of piecewise-linear interpolation between motivation for this problem is the nondestructive digitizaeach pair of successive slices. Our algorithm uses a partial curve tion of objects: after an object is scanned by an echomatching technique for matching parts of the contours, an graphic or an X-ray apparatus, the obtained slices are used optimal triangulation of 3-D polygons for resolving the unfor the reconstruction of the original object. Yet another matched parts, and a minimum spanning tree heuristic for interpolating between nonsimply connected regions. Unlike motivation is the reconstruction of a three-dimensional previous attempts at solving this problem, our algorithm seems model of a terrain from topographic elevation contours. to handle successfully in practice any kind of data. It allows Many solutions were suggested for the pure raster intermultiple contours in each slice, with any hierarchy of contour polation. These usually handle two raster images, where nesting, and avoids the introduction of counterintuitive bridges each pixel is either white or black, or assigned a gray level between contours, proposed in some earlier papers to handle taken from a fixed range. The interpolation produces one interpolation between multiply connected regions. Experimenor more intermediate raster images, which smoothly and tal results on various complex examples, involving actual medilocally turn the first image into the second one. Then, the cal imaging data, are presented and show the good and robust bounding surface is detected using other methods, such as performance of our algorithm.