3D Point Cloud Compression using Invariant Featured Map

Abstract

3D point clouds data obtained from scanning devices require a huge large storage and long post- processing time due to the number of points of the resulting point cloud.  To reduce the processing time of these point clouds and to use them in many computer science application, the simplification is a very important phase.  In our method, we build a graph from the given point cloud then we compute the similarity function using the Menger curvature. We define then a set of patches over the graph and we assign a degree for each vertex depending on its neighbors. We compute next the feature value of each vertex by projecting the point cloud onto different scale spaces and subtract the calculated degree at each scale. Furthermore, we remove iteratively the vertices based on an importance value calculated for each vertex based on the feature value of the vertex and its neighbors. Finally, we present the stability and robustness and of our approach on different 3D point clouds.