Active Contour Modeling For Deformable Surface

Deformable surfaces mean that their shape and topology can be changed under some internal or external forces. They may be elastic surface, plastic surface, inflating surface, shrinking surface, or surfaces obtained by series cutting or wearing. Irregular surface contours, free geometric topological structure and random dynamic changing contours are common characteristics of the surfaces. So it is difficult to establish models for these surfaces with traditional methods. Active contour model was developed based on computer vision theory and is widely used in image processing and graphics modeling. It can combine the lower vision features from image itself with priori information from higher level knowledge understanding, and play an important role in the research field of deformable surface. With regard to the problems and the drawbacks arisen in modeling the deformable surfaces, more attention is paid to the improvement of basic theory of active contour models, and some novel schemes and improved methods are proposed and testified. The main contributions of the dissertation are as following:The gradient vector flow was proved to be an effective way in image segment and boundary extraction. Based on analyzing the gradient vector fields distributed near the contours of deformable surface, we find that they are parallel and opposite in some local areas where the contours are deep concave or bottleneck forms, so there are not any suitable forces to guide active contours moving towards the boundary of interest. Adaptive gradient vector flow algorithm was proposed for boundary extraction as an improved method of the gradient vector flow, in which adjust factors were introduced to improve characters of diffusion field near the boundary firstly, then an additional force was added in the normal direction of the active contour edge. According to the current location, the evolution directions of the curve can be determined in gradient vector flow field. Combined with gradient vector flow, the supposed algorithm can speed up the convergence while maintain its large capture range, it can solve deep concave problems as well as bottleneck problems.Sensitive initial place and poor effective convergence are main problems for parametric active contours evolution. A method of B-spline active contours boundary extraction based on conjugate gradient vector is presented. With human-machine interface, it gives a B-spline curve which is closed to and similar to the objective edge in shape and location based on Z-transform. Rather than evolve deformable curve with every pixel on the curve in conventional method, the proposed method evolves the curve with controlled node points which are more less points than pixel points in numbers, so computational cost should be decreased. Then, an additional conjugate gradient vector force is added on the evolution curve in the gradient vector field, it can make the active contours convergence to the desired edge quickly.Controllable error surface based on discrete wavelet is presented for analyzing geometric components of deformable surface. With discrete wavelet transform and multi-resolution analysis in L~2(R) space, two different filters are applied separately to high and low frequency filter of original surface, then the outline shape and roughness wave can be obtained from the decomposition surface. The error generated during this process is just high frequency component. Instead of gradient field, a deformable boundary extraction scheme based on active contour model and wavelet multi-scale analysis is presented by using wavelet vector field, it can significantly erase the affection of noise and fuzzy image information near the edges, and be sure the convergence of active contours. Surface deformation may be happened in one or more areas, how to describe their dynamic changes remains an unresolved problem up to now. An improved level set method based on Mumford-Shah model is presented for multi-object area contours extraction. Firstly, by using statistical information distributed in image areas as area evolution energy and combining gradient vector field with normal direction of the curves as boundary abstracted fields, a geometric deformable flow field is generated which can drive active contours evolving towards boundary from inside or outside edges. Then the level set function was modified by introducing a adjusted term, it can adjust its values automatically in curve convergence and make level set function be maintained around signed distance function, searching area can be fully covered, the re-initialization number of level set function in convergence iteration would be decreased while the efficiency would be raised.In summary, the theory of active contour models have been studied, some new ideas and methods for extracting deformable surface contours were given in the dissertation. All of these would improve the modeling of deformable surfaces and promote the development of active contour models.