Research And Implement Of The 3D Surface Reconstruction

3D surface reconstruction is an important topic in the areas of computer vision, pattern recognition and 3D data visualization. The 3D model of reconstruction is also the premise of analysis, emulation and decision-making. The paper focuses on 3D surface reconstruction from slice-images data, the problem can be broken into four subproblems: the correspondence problem, the tiling problem, the branching problem, and the surface-fitting problem.Firstly, a new solution to the problem of contour tiling is proposed in this thesis, based on simulated annealing genetic algorithm, which effectively implements contour tiling in 3D surface reconstruction. This algorithm use the weighted average of internal angle of triangles, area and perimeter as its optimized objective function, and by adding simulated annealing process to traditional genetic algorithm, it improves the diversity of population, avoiding the premature convergence problem existed in genetic algorithms, and effectively enhancing the ability to searching for the global optimum. Because cross operation costs most of the time in the algorithm, a new approach of improving the efficiency of the cross function is described, using a new minimum crossover polygon algorithm based on edge. The experimental results show that the new cross algorithm remarkably improves the efficiency of the cross function. Special data structure is designed, according to the characteristic of contours tiling problem, to improve the efficiency of the algorithm. And also, to deal with different situation of contour tiling, a method of tilling contours through adding accessorial line manually is implemented.Secondly, to deal with the branching problem in 3D surface reconstruction, a new method that can effectively handle the branching problem is proposed. The algorithm applies Delaunay triangulation to branching problem according to the characteristic of the Delaunay triangulation of constrained edge point sets in plane. This method projects the neighboring section contours onto a same section which creates a constrained edge point set. The constrained edge point set in plane is Delaunay triangulated, then new contours are created according to the triangulation, and it forms one-to-one relation of the contours. The experimental results indicate that the algorithm performs well in practice, and can deal with of different situation's effectively. A method is described, which can handle flexibly kinds of branching problem by manually decompose contour.Finally, the classification and application of the 3D model and the visualization theory in scientific computing are introduced. The algorithms are implemented in practical project, and attain rather satisfactory result in experiment and in practice.