Contour Extraction And Surface Reconstruction Based On Deformable Models

Contour extraction and surface reconstruction is an important problem in computer vision, and can be used extensively in many fields such as virtual reality, autonomous guided vehicles, robot environment analysis, object tracking and recognition in monitor system, biology medical image processing, industry online automatic checking and reverse engineering etc. 3D reconstruction based on layer data has been a hot point of research and new direction of reverse engineering. The contours extracted from cross-sections can be used to surface reconstruction to obtain the object's CAD model, or can be used directly to prototyping system. During this process, contour extraction form cross-section is the key step.In the last few years, introducing physics principle into shape recovery has attracted researchers' attention, and deformable model based contour extraction and shape reconstruction is one kind. A deformable model is an energy-minimizing spline model controlled by internal forces and external forces. Internal forces come from geometry model which constrain its shape, and external forces come from image data which guide its movement and pull it toward image feature. Deformable models which bring a new viewpoint to traditional computer vision have been successfully applied to many fields in computer vision such as image segmentation, motion tracking, 3D reconstruction and stereo matching etc., and have become one of the most active and successful fields in computer vision and pattern recognition.Therefore, cross-sectional contour extraction and surface reconstruction based on deformable models is important signification of theory and application. Aim at medical cross-sections, the research work of this dissertation put emphasis on B-spline deformable model. In this dissertation, based on the systematically analysis of theory and application research on deformable model, a new B-spline active contour based on the finite element method (FEM) is proposed; then B-spline active contour is applied to fitting the digital curve cross-sectional contour, and adaptive FEM is adopted to improve the precision; lastly the model is generalized to 3D, and a B-spline active surface model based on FEM is proposed. The main work and conclusions in this dissertation are as follows:(1) The basic theory of deformable model is expatiated and the physical essence of deformable model is discussed from the viewpoint of elasticity theory. The research, development and application of deformable model is reviewed from geometry representation, energy function and optimization method, and the existing problems and possible future research orientations are presented. (2) A B-spline active contour model based on FEM is presented. B-spline is the most popular method to represent free-form curve and surface, and FEM provides an efficient way to solve the functional minima problem. In this model, a cubic B-spline curve segment is used as one element, and the FEM is adopted to find the B-spline active contour which minimize its energy. Experiment results showes that this medel could effectively combine the merits of B-spline and EFM for active contour model, yielding stable, accurate and faster convergence, and its result is favorable for surface reconstruction.(3) A B-spline active contour model based on adaptive FEM is presented and applied to surface reconstruction from layer data. Firstly edge detection method for cross-sections is expatiated, and a B-spline active contour based on adaptive FEM is proposed to fit the digital contour curve. This method can avoid the sampling and parametrization problems result from traditional interpolation and approximation methods and improve fitting accuracy is gradually improved during fitting process by insert new control points to realize the satisfying result. Lastly 3D model is reconstructed from a set of planar B-spline contour. (4) To generalize the B-spline active contour to 3D, a B-spline active surface model based on FEM is proposed. By anchoring an imaginary spring between each data p...