Research On 3D Surface Digital Patching Algorithms

With the rise of Digital Museum in 1990s, cultural relics has been converted into digital form for collection, management, display and handle, all of these provides a new idea for cultural heritage custody. The development of computer graphics, pattern recognition and 3D information processing makes it possible to patch cultural debris with rich characteristics and fine textures in digital form. Digital patching has many advantages, such as high speed, re-operation, avoiding inadvertent damage. Moreover, this method can also be used for the cultural debris repair which can not be handled with hands and provides a new solution for the preservation of cultural debris. However, the existing patching algorithms are almost based on 2D shape features, and there are still some unsolved problems and disadventages to be studied. Considering 3D data of cultural debris'surface as the research object, the aim of this dissertation is to develop a set of completed and effective 3D surface digital patching algorithm.This dissertation systematically summarizes the current achievement of 3D digital patching algorithms and the fundamental theory of pattern recognition and computer graphics. To improve the accuracy of cultural debris classification and cultural debris patching algorithm, considering that 3D laser scanning data contains large amount of data and has complex topology, this dissertation has some researches on automatic cultural debris classification, 3D model surface feature extraction, 3D model surface region segmentation and 3D digital patching algorithm.In the automatic cultural debris classification, considering that nowadays there is no automatic classification algorithm for cultural debris with different materials, different types and different ages, this dissertation proposes three automatic cultural debris classification algorithms based on surface features. With 2D texture as surface feature, text regions in digital images are detected and located using text extraction algorithm based on Gabor analysis, and the cultural debris with text and those without text can be automatic classified. With 3D texture as surface feature, cultural debris is automatic classified using SVM as classifier, 3D texture histogram model and 3D texture correlation function as classification feature. Simulations show that the above methods have effective and automatic classification for cultural debris.In the 3D model surface feature extraction, considering that there are few surface feature extraction algorithms for 3D point cloud model and the performance of current 3D mesh model feature extraction algorithm is influenced by sampling and voxelization, vector-based descriptor for 3D point cloud model and spherical grid D2 descriptor for 3D mesh model is proposed. In vector-based descriptor for 3D point cloud model, points in point cloud are precisely denoted by a set of vectors, some statistics of the vector set are computed as the descriptor for 3D point cloud model. Based on analysis of advantage and disadvantage of classic D2 and GD2 descriptor, this dissertation proposes spherical grid D2 descriptor for 3D mesh model. Spherical grid is chosen as the replacement of the random points in D2 and the 3D Cartesian coordinate grid in GD2, the proposed descriptor outperforms other methods.In the 3D model surface region segmentation, considering that surface coarseness influences the performance of current 3D surface region segmentation and current algorithms can only handle the debris with regular shape, flat surface and few broken surfaces, a surface segmentation method for 3D laser scanning data based on region dilation strategy is proposed, and this method segments the 3D laser scanning data into distinct region of similar orientation. Firstly, 3D point clouds produced by 3D laser scanner are converted into polygon mesh. Then, normal vectors and areas of every polygon mesh are computed and some candidate regions are obtained using region dilation strategy. Finally, the 3D laser scanning data is segmented into regions of similar orientation after the elimination of the insignificant regions. The performance of proposed method is testified using 3D laser scanning data of actual objects, and the experimental results show that this algorithm works well on various types of 3D laser scanning data.In the 3D digital patching algorithm, contraposing the phenomenas of that all the current algorithms use curves of debris as match features and there is no algorithm to take full advantage of 3D shape features, this dissertation proposes 3D digital patching method based on height-map of debris broken surface. The 3D data of debris surface is segmented based on region dilation strategy firstly. Then the broken regions are determined via surface coarseness. The global-best patching scheme is achieved by employing height-map as matching feature and complementariness between height-map as matching performance functional. The performance of proposed method is finally verified by simulations.