Research On Theory And Algorithms Of 3D Reconstruction From Slice Images

3D reconstruction from slice-image is a main topic of visualization of 3D volume data. This problem has broad applications in geology and aerograph, especially in medicine. With the development of computer technology and computer graphics, 3D reconstruction based on medical slice-images has became a new step of the computer technology. This dissertation mainly studies and deals with the technique of 3D reconstruction from cross-sectional contours and some related techniques.The technique of 3D reconstruction from cross-sectional contours mostly comprise pre-process, segment, contours extraction and surface reconstruction. Surface reconstruction can be decomposed into four sub-problems, the correspondence problem, the branching problem, the tiling problem and the surface fitting problem. This dissertation gives a survey of previous research on related basic algorithms. Then, some improves algorithms are given.Because of the data redundancy, a new condensation method for contour data has been proposed. Besides, periodic B-spline fitting method is adopted to approximate a group of new contour data. It is a curve that can figure the contour of the original CT image.In the process of linking triangle, an improved algorithm based on the shortest diagonal line is advanced. By judging the distance between next two points and the midpoint of present edge, the shorter edge should be selected as the next triangle edge. This method is purport if shape difference of adjacent contours is more.The solution to the correspondence problem and branching problem is important for 3D reconstruction of surface from contours, especially for validity of topology and geometry of reconstructed triangular mesh. By considered the ratio of center distance and layer distance, and cover area between contours of adjacent slices, the correspondence coefficient is proposed as geometrical constraint. If the correspondence coefficient between contours is larger than a given threshold value, the correspondence relations are defined. Then the rule of corresponden- ce validity is used to pledge the validity of topology. Besides, a method based on linear interpolation is presented for solving branching problems. This method is applied to solving reconstruction problem of adjacent contours which curvature change is less. Geometry shape of 3D reconstruction model could be hold.Surface fitting is another method of 3D reconstruction from cross-sectional contours. By the study of the B-spline interpolation method, a new ascertain method about common node vector is presented, which is an approximate surface for original contour data. Applying this method to a group of single contour data, 3D reconstruction surface is perfect.