3D object reassembly using region-pair-relation and balanced cluster tree. (December 2020)

• Record Type: Journal Article
• Authors: Liao, Sheng-hui; Xiong, Chao; Liu, Shu; Zhang, Ying-qi; Peng, Chun-lin
• Is Part Of: Computer methods and programs in biomedicine. Volume 197(2020)
• Publication Date: 2020-12
• Subjects: Object reassembly -- Region-pair-relation -- Coarse-to-fine matching -- Fracture reduction; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2020.105756
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: A generalized framework for 3D object reassembly is proposed. A novel region-pair-relation descriptor with enhanced geometric constraints is introduced to ensure the algorithm accuracy and efficiency. A balanced cluster tree is designed to remove the outliers and accelerate the spatial decomposition. semi-interactive manipulation is provided to deal with the complex objects. Abstract: Background and Objective: Object reassembly is a key technology in scenarios such as surgical planning and broken object restoration. Based on previous research, this work intends to explore the general tasks of 3D object reassembly, including conventional object reconstruction and bone fracture reduction. Methods: We introduce an efficient and robust region-pair-relation descriptor, which incorporates strong geometric constraints and remains invariant to rotation and translation. We segment the fractured objects using balanced cluster tree, and develop a coarse-to-fine method for object reassembly. The matching quality of potential region contact pairs at different depths is estimated recursively from the root of the tree. Once the best contact pairs are determined, the least squares method is implemented to obtain the matching results. In addition, we also provide a semi-interactive manipulation to deal with the complex objects. Results: For most types of broken objects, our approach can generate high accuracy matching results within 10 s, with the cluster tree depth equals to 11. It allows the automatic reassembly of different-sized fragments. For bone fracture blocks with cancellous structures, a semi-interactive operation is integrated so that the precise matching can also be achieved in 30 s. Conclusion: The proposed framework can be expanded to various object reassembly tasks in either automated or semi-automated manner, including the fracture reduction problem which used to be an intensive manual process. Therefore, our work shows significant advantages in medical applications.