Denoising method based on multiscale distribution score for point cloud

Patent metadata

Field	Value
Publication number	US-12469113-B2
Application number	US-202318366604-A
Country	US
Kind code	B2
Filing date	Aug 7, 2023
Priority date	Mar 1, 2023
Publication date	Nov 11, 2025
Grant date	Nov 11, 2025

Abstract

A denoising method based on a multiscale distribution score for a point cloud includes: constructing a two-layer network model based on multiscale perturbation and point cloud distribution, where the two-layer network model includes a feature extraction module for extracting a feature of the point cloud and a displacement prediction module for predicting a displacement of a noise point; constructing a point cloud noise model for improving a denoising effect and retaining a sharp feature and avoiding reducing quality of point cloud data; extracting a global feature h by inputting the point cloud data into the feature extraction module; iteratively learning the displacement of the noise point by the displacement prediction module according to a feature obtained by the feature extraction unit; and defining a loss function of network training, and completing convergence under the condition that the loss function reaches a set threshold or a maximum number of iterations.

First claim

What is claimed is:

1 . A denoising method based on a multiscale distribution score for a point cloud, comprises:
step 1: constructing a two-layer network model, wherein the two-layer network model comprises a feature extraction module for extracting a feature of the point cloud and a displacement prediction module for predicting a displacement of a noise point; step 2: constructing a point cloud noise model for improving a denoising effect and retaining a sharp feature and avoiding reducing quality of point cloud data; step 3: extracting a global feature h by inputting the point cloud data into the feature extraction module, wherein preprocessing the point cloud data, enhancing an anti-noise performance of a network by adding multiscale noise perturbation to processed point cloud data, and extracting, with Encoder, the global feature h of the point cloud by the feature extraction module; step 4: iteratively learning the displacement of the noise point by the displacement prediction module according to a feature obtained by the feature extraction module; and step 5: defining a loss function of network training, and completing convergence in response to the loss function reaches a set threshold or a maximum number of iterations; wherein the step 1 further comprises: preprocessing a neighborhood of an input noisy point cloud by the feature extraction module, and the anti-noise performance of the network is enhanced through the multiscale noise perturbation; wherein a displacement estimation module of the displacement prediction module obtains a distribution score of a neighborhood point cloud according to a score estimation unit, considers a position of each point, further covers a neighborhood of the point, and finally completes a denoising process by iteratively learning the displacement of the noise point; wherein wherein the neighborhood point cloud refers to a set of data that have a distance less than a specific distance from a selected point in current point cloud data; wherein the point cloud distribution refers to that point clouds scattered in a certain area obey a distribution function, wherein the distribution function shows statistical regularity of a random point cloud; wherein the multiscale noise perturbation refers to use of multiscale isotropic Gaussian noise with a mean value of 0 to interfere with the data.

2 . The denoising method based on the multiscale distribution score for the point cloud according to claim 1 , wherein the step 2 further comprises:
step (2.1), regarding in the present invention a noise-free point cloud Y={y i } i=1 M as a set of samples p(y) of three-dimensional distribution p supported by a two-dimensional manifold, deducing p(y)→∞ in response to the noise point y is just on the two-dimensional manifold, assuming that noise follows distribution n, and modeling the noisy point cloud X={x i } i=1 M as shown in the following formula to reduce the number of point clouds in the denoising process:

X
=

{

x
i

=

y
i

+

n
i

}

i
=
1

M

wherein M represents the number of point clouds, n i represents a component of noise distribution n, x i represents a component of a noisy point cloud X, and y i represents a component of s noise-free point cloud Y;
wherein the two-dimensional manifold refers to a compact topological space in a two-dimensional space, and each point in the topological space is an interior point; and
step (2.2), representing a probability density function q(x) of the noisy point cloud X as a convolution (p*n)(x) between a point cloud distribution p and the noise distribution n, and simultaneously taking derivatives of both distributions, wherein in response to n equals 0, a noise-free point cloud Y from a noise-free distribution p is just located at q:

q
⁡
(
x
)

:=

(

p
*
n

)

⁢

(
x
)

=

∫

p
⁡
(
y
)

⁢

n
⁡
(

x
-
y

)

⁢
dy

wherein the probability density function of the noisy point cloud represents a probability that the noise point falls within a certain specified range;
wherein the convolution is a mathematical operation to generate a third function from two functions, and the convolution is a special integral transformation.

3 . The denoising method based on the multiscale distribution score for the point cloud according to claim 1 , wherein the step 3 further comprises:
step (3.1), preprocessing collected point cloud data to make a format directly processed by a neural network; step (3.2), computing a rotation matrix with principal component analysis (PCA), aligning a point cloud; step (3.3), obtaining x σ i by adding multiscale noise perturbation to an input point cloud x, and processing data with perturbation signals separately, wherein an output of the network is a weighted result of different noise scale processing; and step (3.4), overcoming limitation of a linear mode by adding several hidden layers, extracting the point cloud feature by the network through mapping the data to different dimensions through multi-layer perceptron (MLP) of shared parameters, and obtaining a potential feature of the point cloud through a convolution.

4 . The denoising method based on the multiscale distribution score for the point cloud according to claim 3 , wherein the step (3.1) further comprises:
step (3.1.1), considering that a denoising problem of the point cloud is regarded as a local problem, a denoising result of any noise point x i comes from a local neighborhood {tilde over (X)} of the point, and a distance between {tilde over (X)} and x i does not exceed a given neighborhood radius r:

X
~

=

{

x

Assignees

• Univ Jiliang China
• Univ Zhejiang Technology

Inventors

• Xiao Gang
• Lu Jiawei
• Wang Qibing
• Li Chen
• Hu Hao
• Yao Junbo

Classifications

• G06T5/60 Primary

  using machine learning, e.g. neural networks · CPC title

• G06T2207/10028

  Range image; Depth image; 3D point clouds · CPC title

• G06T5/70 Primary

  Denoising; Smoothing · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title