Studies On Feature Extraction And Hole-Filling In3D Digital Geometry Processing

Nowadays, the society has entered into the period of information, digital is a distinct char-acteristic of this period. Information gathering and expression, reconstruction and other digi-tal processing techniques bring a large number of new topics for basic researching. With the rapid development of three-dimensional scanning technology, three-dimensional digital geom-etry model has become an emerging digital media and has been used in widely ranges, which also brings out many opportunities and challenges for three-dimensional digital geometry pro-cessing. This dissertation does some researches on point clouds consolidation, feature extraction and hole-filling in digital geometry processing. The main works are listed as follows:(1) We present a new consolidation method for unorganized point clouds that are frequent-ly corrupted in the existence of noises, outliers, and thickness due to multiple scan fusion. Our novel method is primarily motivated by the concept of principal surface, which is the smooth surface passing through the middle of the data. At the beginning of the algorithm, we con-struct the local neighborhood of a point by utilizing the shared nearest neighbor relationships. Then, according to the self-consistent condition of principal surface, we propose a feature-aware projection operator for updating the principal surface to approximate the underlying model. Our method not only incorporates the denoising process into the framework of point clouds thinning, but also gets the thinning results with sharp and geometry feature preserving.(2) To detect sharp features on scanned point clouds, we propose a novel feature extrac-tion method based on local reconstruction and subneighborhood selection. This method first constructs the triangle sets in the local neighborhood of the initial feature vertex, which can reflect the feature structures of the underlying model. Then, the triangle sets can be clustered into different sets according to the normal distance between them, which indicates the clarifi-cations of points in the local neighborhood. Finally, we can recognize the real sharp feature vertex by checking whether the current point lying on the intersection of multiple fitting plans of sub-neighborhoods or not. Comparing with the existing techniques, our method is robust to noises in low complexity and can effectively distinguish the pseudo feature points from the real features, which will provide sufficient information for feature-aware point clouds processing. (3) Focusing on the feature extraction on triangle meshes, we propose a sharp feature ex-traction method based on the idea of neighborhood supporting. A novel salience metric based on the idea of neighborhood supporting is defined, which can promote the values of real features and suppress the values of noisy points to some extent. The novel defined salience metric plays an important role in feature extraction on triangle meshes. Different to the existing methods, the advantages of our method are that in the process of sharp feature extraction, it is not only robust to noises, but also can extraction as many weak features as possible.(4) To complete the big holes in curved regions, we propose an effective and simple hole-filling technique based on the advancing front method. After detecting the boundary of the hole, we first estimate the reasonable normal of each boundary vertex and classify them into concave and convex. According to these information, we adjust the vertex inserting direction and compute optimal vertex position to create new triangle to restore the missing shapes of the hole. Without any triangle optimization, the new created triangle mesh has consistent mesh distribution and smooth transition with the existing meshes. Additionally, the quality of the new generated triangle reaches a high level.(5) To recover the missing sharp features in the hole region, we propose an automatic feature-preserving hole-filling method. By defining the diversity between boundary sharp fea-ture vertices and concave and convex analysis, we automatic achieve the boundary sharp feature vertices matching and missing feature lines reconstruction, which avoid the time-consuming manual operation and improve the efficiency of the algorithm. The missing feature structures are explicitly constructed by cubic splines, which interpolate the corresponding feature vertices and guarantee the accuracy of the recovered feature structures. Additionally, we predict the position of missing corners by minimizing a novel defined energy and successfully recover the missing corners in our feature-preserving hole-filling algorithm.