Point Cloud Denoising Based on Improved Maximum Likelihood Method
TANG Yuqin
CAI Yong
ZHANG Jiansheng
Abstract：As a common geometric data type,point cloud contains abundant geometric space information.However,the point cloud data obtained by acquisition equipment or image reconstruction is often disturbed by noise.It wills affect the downstream tasks of point cloud processing.Combined with the optimization idea in deep learning,this paper improves the network structure of this method,links hierarchical features,freezes the feature extractor and retrains the score estimation unit to improve the performance of the network.To better protect the characteristics of the point cloud,this paper optimizes the loss function of the network and intro-duces the regularization term,so that the points are regularly distributed on the surface.The experimental results show that the chamfer distance of the point cloud after denoising is reduced by 7.6%compared with the original network,and the denoising effect is effectively improved.The denoising effect of the improved algorithm is not only superior to state-of-the-art deep-learning-based denoisers,but also superior to state-of-the-art optimization-based denoisers.
Keywords：deep learningpoint cloud denoisingmaximum likelihood methodfeature protect
Publication Date：2025-04-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：6( 936-941 )