Study On Techniques For MRI Based Surfaces And Solids Reconstruction Of Fruits

Three dimensional (3D) geometric models are the modeling basis of virtual reality and scientific visualization. Voxel rendering and surface rendering techniques are developed to reconstruct 3D surface and solid models from magnetic resonance images (MRI). This work focus on the surface based method. Models reconstructed are expected to be used in fields like the detection of inner pathological deterioration of fruits, the fruit simulation, and the horticultural exhibitions for teaching purpose.Standing on the shoulders of classical methods like Delaunay triangulation and Advancing Fronts, existed reconstruction systems create high-quality meshes which are able to meet the requirements of visualization and finite element method (FEM). Nevertheless, most of them are suffering from two open problems. One is the computational burden of calculation of Voronoi diagram and Steiner points; the other is that the versatility of reconstruction systems is reduced since the inputs are confined to be some special topologies. These problems inspired us to explorer elegant reconstruction algorithms with lower computational consumption and enough independence with respect to topologies of inputs.In this thesis, vital techniques of contour-based reconstruction, including the contour extraction, the contour-based complex surface models reconstruction, and the contour-based solid model reconstruction are studied to extract 3D surface and solid models from MRI images of fruits. In comparison with point clouds and piecewise linear complexes, contours contribute better border constrains. Contour extraction extracts contours from MRI image sequences and provides data for the reconstruction process. It is an influential step with respect to the accuracy of models reconstructed. Structural contours are extracted from MRI images employing APIs in the OpenCV library. Afterwards two methods are implemented to sample key points to generate discrete contours with reasonable density of vertices. The surface model reconstruction is split into two passes including contours matching and titling. Based on the distance and the similarity of different contours, a contour matching algorithm is proposed to tackle the probably emerged ambiguities during correspondence, and to decompose contours lying on neighboring slices into contour pairs. Then the contour pairs are tiled using the improved simple-constrained triangulation algorithm and triangular surface meshes are made. The refined tiling algorithm is able to process complex contour pairs with nests, branches and projective intersections. G1 continuity surfaces are obtained by interpolating the rebuilt surface meshes with a new triangular Gregory patch. The tetrahedral solid models of fruits are produced by the Tetgen library.The high computational consumption of current tetrahedralization systems as well as the rigid input constrains of them inspired the exploration of simple constrained solid reconstruction method in our work. We designed a prototype of simple constrained solid tetrahedralization that creates surface and solid models simultaneously without inserting Steiner points. Further study is needed to enrich the prototype.