A Novel Medical 3D Reconstruction Based on 3D Scale-Invariant Feature Transform Descriptor and Quaternion-Iterative Closest Point Algorithm

With the development of modern medical technology, 3D reconstruction technology has been more and more used in disease diagnosis and theoretical medical research. In the process of reconstructing medical 3D model, this paper proposes a novel medical 3D reconstruction algorithm based on "coarse-to-fine" registration. Firstly, a random sample consensus (RANSAC) coarse registration algorithm based on key points is proposed. The key points in point cloud are extracted by 3D scale-invariant feature transform (3D-SIFT) descriptors, which narrows the scope of point selection in the coarse registration, thus the computational cost is significantly reduced. Secondly, we propose an novel iterative closest point (ICP) fine registration algorithm, which is based on quaternion algorithm (Quaternion-ICP). The initial coordinate transformation of ICP algorithm is set with the result of coarse registration, and the corresponding points between two point clouds are identified by Kd-tree algorithm. The quaternion algorithm is used to solve the rotation matrix and translation vector, which has a high execution efficiency and there are no singular points in describing the rotation matrix. Compared with other algorithms, the experimental results show that our novel medical 3D reconstruction algorithm is significantly superior to several representative methods, and improves the registration efficiency and accuracy.