Towards uniform point distribution in feature-preserving point cloud filtering

Point cloud filtering must balance two competing objectives: suppressing noise while keeping sharp edges and corners intact. Methods that prioritise feature preservation typically allow point positions to cluster near high-curvature regions, producing non-uniform densities that degrade downstream applications such as surface reconstruction, rendering, and registration. This paper makes the observation that uniformity and feature preservation can be simultaneously achieved by decoupling the two objectives into complementary optimisation terms. A normal-guided feature detection step classifies each point as belonging to a sharp feature, a flat region, or a smooth curve, and assigns an appropriate filtering kernel. A spatial regularisation term then repels neighbouring points that have drifted too close together while attracting those that are too sparse, independently of the feature labels. The joint optimisation converges to filtered point clouds that are both feature-sharp and spatially uniform. Validated on ShapeNet and scanned real-world objects and published in Computational Visual Media (2023), the method outperforms dedicated feature-preserving and uniformity-enforcing baselines in isolation, and is particularly beneficial as a pre-processing step before Poisson surface reconstruction.

Problem setting

Feature-preserving point cloud filtering removes acquisition noise while retaining sharp geometric features such as edges and corners. However, most existing methods do not control the spatial distribution of filtered points, often producing uneven sampling densities: points cluster near sharp features while flat regions become sparse. This paper proposes a filtering approach that jointly optimises for feature preservation and uniform spatial distribution.

In the broader publication record, this work appears in Computational Visual Media, 9(2):249–263. The visual notes below pair the paper’s original figures with a concise reading of the method, experimental setup, and reported results.

Method and visual evidence

The method works on 3D geometric observations such as point clouds, poses, correspondences, or segmented regions, then uses the proposed representation to improve robustness under noise, viewpoint change, or limited observations.

The extracted figures below show the geometric representation, network or optimization pipeline, and qualitative or quantitative results.

Method overview. This image is extracted from an embedded PDF image object on page 6, then recomposed for web display.

Representation and setup. This image is extracted from an embedded PDF image object on page 6, then recomposed for web display.

Experimental evidence. This image is extracted from an embedded PDF image object on page 7, then recomposed for web display.

Result comparison. This image is extracted from an embedded PDF image object on page 7, then recomposed for web display.

Additional visual result. This image is extracted from an embedded PDF image object on page 8, then recomposed for web display.

Results and impact

The evaluation reported in Computational Visual Media, 9(2):249–263 uses the extracted figures above to show the method’s measurement, reconstruction, segmentation, matching, or diagnostic behavior on representative experiments. These visuals are paired with the paper’s quantitative or qualitative analysis to make the workflow easier to inspect from the homepage.

Source handling

I extracted 23 candidate image objects from paper.pdf and generated the compressed WebP figures used on this page. The local PDF was also optimized from 979,871 bytes to 976,013 bytes.