Extended Gaussian sphere and similarity fusion method for reassembly of 3D cultural relics

Screening of the best matching pair of broken pieces of 3D cultural relics is an important step in realizing their automatic reassembly. Herein, a reassembly framework based on an extended Gaussian sphere (EGS) and a similarity function is proposed. First, the inner surface was identified based on the proportion of the point cloud and the smoothness of the broken pieces, thereby avoiding over-segmentation. The EGS model was then used to represent the geometric and color information of the inner surface boundary. The similarity function, which was defined to describe the global characteristics of the EGS model, can help the program to automatically screen the two most similar broken pieces to be matched. In this way, the rough alignment step was completed. Finally, a fine alignment step was conducted using the iterative closest point method. This reassembly framework is effective for broken bowl-shaped pieces. The average screening accuracy of our final method is 93.87% and the calculation time is 9.7 s. Compared with a traditional Gaussian sphere model based on a normal vector and curvature, the proposed method can improve the screening accuracy and average calculation speed by 6.60% and 16.08%, respectively. Experimental results with real datasets demonstrated the validity and effectiveness of the proposed method.