Balanced feature-sensitive point sampling for 3D model generation

Abstract

Many surfaces used in practice consist of combinations of various features that are sufficiently smooth or highly detailed. In the presence of highly detailed features, it is obvious that non-feature-sensitive sampling is unable to preserve small features after significant data reduction. The higher the sensibility of point sampling to small features, the more intensive sample points in detailed features we have. When feature sensitivity is exaggerated, however, it makes the distribution of sample points locally greedy and unbalanced, which is undesirable for many point-based applications. Therefore, it is important for feature-sensitive sampling to blend feature sensitivity with balanced distribution of sample points to generate high-quality 3D models through point sampling. This paper presents a novel 3D point-sampling algorithm that produces balanced feature-sensitive sample points from unorganized point clouds. The key ideas of our point-sampling algorithm are that we set scalable feature sensitivity based on local feature size and balance the distribution of sample points by normalizing the number of sample points over localized regions. Then, sample points are placed at feature-sensitively weighted positions by a fuzzy clustering technique. The proposed sampling method enables us to produce a variety of well-balanced sample points from sparse to dense and from regular to feature-sensitive. We demonstrate the versatility of our sampling method by providing experimental results and comparisons with other point sampling methods.