An Online Rapid Mesh Segmentation Method Based on an Online Sequential Extreme Learning Machine

Abstract

The existing mesh segmentation methods currently require long training times and have high computational complexity. Consequently, many of these methods cannot meet the rapid requirements of digital geometry processing in the Web environment. This paper proposes an online rapid mesh segmentation method based on an online sequential extreme learning machine (OS-ELM). In the training stage, the OS-ELM is trained by analyzing the mapping relationship between the shape descriptors of the mesh and the Gaussian curvature threshold. We reduce the dimensionality of the shape descriptor vector via principal component analysis (PCA) and extract the Gaussian curvature threshold of the mesh as the sample label using statistics. In the segmentation stage, the Gaussian curvature threshold is quickly extracted to realize the online rapid mesh segmentation via the OS-ELM. Simultaneously, the OS-ELM is updated to realize online incremental learning based on a small number of training samples. Our method is verified using the meshes provided from ShapeNetCore. The experimental results indicate that segmentation results similar to manual segmentation can be rapidly generated online using our method.

Keywords

• Online sequential extreme learning machine
• rapid mesh segmentation
• incremental learning
• Gaussian curvature threshold
• web environment