Mapping bootstrap error for bilateral smoothing on point set

Abstract

In this study, point set smoothing on surface reconstruction is developed. In surface reconstruction, the objective is to find the best surface representations of a modeled object. Surface representations of 3D model are created from 3D point sets. The raw data from optical devices such as laser scanner always contain noise. The surface reconstruction process should yield consideration to this data noise for a better representation of the modeled object. Surface reconstruction from sample points is a difficult challenge and the problem arise in application such as medical imaging, visualization and reverse engineering. Bootstrap is a model averaging technique and widely used in applications such as noise estimation. We used the implementation of bootstrap error estimates on point sets to perform smoothing on noisy models. Given a noisy model of a point cloud, error is estimated on each point using bootstrap error estimation. Then, a surface representation is reconstructed from the set of point clouds. We model a function or algorithm that could adapt the density of the model based on bootstrap error estimation to avoid oversmoothing, irregular noise on the data and feature preservation. To smooth the model, we use the concept of bilateral filtering found in noise filtering of 2D image. We compare the curve fitting in linear mapping and quadratic mapping in which the quadratic mappings have two models. In this paper, we used bootstrap error estimates to guide a projection based on point set smoothing algorithm. As a result, the noisy points were smooth out and our model had been recovered. We expect that bilateral smoothing would avoid over smoothing the feature areas and feature of the model preserved.