Denoising for surface reconstruction

Abstract

We present an algorithm to denoise an unorganized point cloud which contains noise, white noise and outliers for surface reconstruction. Our algorithm first removes the outliers, white noise and very noisy points based on the point cloud behavior and simple statistical analysis. Afterwards, our algorithm denoises the remaining points by a modified Laplacian smoothing algorithm. Finally, we apply the Robust Cocone algorithm to reconstruct the surface from the denoised point cloud followed by some postprocessing on the reconstructed surface for constructing the boundaries and further smoothing. Our algorithm applies a variant of the standard octree structure to manipulate the points and this makes our algorithm very efficient. The experimental results show that our algorithm can generate a smooth surface even though the noise level is very high and the running time is fast. The data sets that we experimented with include some raw data contaminated by artificial noise, white noise and clusters of outliers. We also compare our algorithm with Robust Cocone, the Adaptive Moving Least-Squares (AMLS) algorithm, and the PoissonPU algorithm. The comparison results show that our algorithm can improve the surface reconstructed from Robust Cocone algorithm, and it gives better results than the AMLS algorithm and the PoissonPU algorithm.

Date of Award	2012
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology