skeletal model de-nosing with cascaded network

With the continuous breakthrough of three-dimensional(3D) data visualization and 3D printing technology, there are many achievements in the combination field of computer science and medical science. Especially in orthopedics, the 3D printed skeletal model and prosthesis have a good effect on surgeon's surgery and patient rehabilitation process. The 3D printed skeletal model plays a vital role in the surgical process, and the accuracy of the three-dimensional printed skeletal model determines the extent of this auxiliary effect. In the field of orthopedics, we propose to use the data-driven the radial basis function(RBF) network to remove noise from the surface of the skeletal model. The RBF network achieves the goals of denoising by learning the regression relationship between the noisy mesh and the geometry descriptor called the filtered facet normal descriptor (FND) based on bilateral normal filters of the corresponding smooth facet, without making many unreasonable assumptions made by the traditional denoising algorithms. Based on the cascaded model, we adjusts the noise level on the training model according to the real situation of the bone reconstruction model, and changes the structure of the geometric descriptor composed of the three-layer cascaded regression function network. The cascaded regression model suitable for skeletal three-dimensional model denoising is trained while maintaining the mesh mean normal vector deviation angle of 4.2°. The model not only removes the noise of the bone 3D model, but also maintains the geometric characteristics of the skeleton model itself.