Saliency-Guide Simplification for Point-Cloud Geometry

Abstract

To efficiently simplify large-scale point clouds and keep geometric details as many as possible, we propose a novel operator guided by point-saliency. Firstly, we adopt a site entropy rate algorithm to calculate the saliency value which represents the significance of every point. Intuitively, the point of higher value should be retained. We introduce the saliency value as a weight term to locally optical projection (LOP) operator. What's more, we incorporate locally adaptive density weight into our operator to deal with the highly non-uniformed point clouds. Compared with other methods, our approach preserves more spatial information when down sample a point cloud to a certain number of points. Experimental results also show that our method is highly robust to noise and outliers.