基于可逆神经网络的点云几何有损编码.
Gespeichert in:
| Titel: | 基于可逆神经网络的点云几何有损编码. (Chinese) |
|---|---|
| Alternate Title: | Lossy Point Cloud Geometry Encoding Based on Invertible Neural Network. (English) |
| Autoren: | 王楷元, 方志军 |
| Quelle: | Radio Communications Technology; 2025, Vol. 51 Issue 6, p1351-1358, 8p |
| Abstract (English): | With the rapid development of wireless sensing technology and laser digital acquisition technology 3D point cloud data has been widely used in multiple fields. However the large scale and high redundancy of point cloud data make great challenges to its application and the industry urgently needs efficient point cloud geometric lossy encoding algorithms. Traditional point cloud geometric lossy encoding algorithms have low efficiency and poor encoding performance while deep-learning-based point cloud geometric encoding algorithms mostly use AutoEncoder AE neural network architectures which suffer from a certain degree of feature information loss. In addition in recent years most studies have focused on improving the entropy encoding while neglecting the optimization of the transformation between point cloud geometry space and its potential feature space. In response to the above issues this paper proposes a lossy point cloud geometry encoding algorithm based on Invertible Neural Network INN . This algorithm uses an INN with mathematically rigorous reversible properties for feature extraction of point cloud geometry information avoiding information loss during the encoding process and ensuring the stability of reconstructed point clouds during the decoding process. This paper designs a 3D-Dense-Block module and a channel squeeze module to enhance feature information increase the non-linear expression ability of the algorithm network and avoid the occurrence of suboptimal solutions during training. Experimental results show that the algorithm achieves better rate distortion performance than the Moving Picture Experts Group MPEG benchmark algorithm on Microsoft Voxelized Upper Bodies MVUB and MPEG 8i datasets. [ABSTRACT FROM AUTHOR] |
| Abstract (Chinese): | 随着无线电传感技术与激光数字采集技术的快速发展, 3D 点云数据在多个领域得到广泛应用。 然而, 点 云数据规模庞大、冗余度高, 给其应用带来巨大挑战, 业界亟需高效的点云几何有损编码算法。 传统点云几何有损编 码算法效率较低、编码性能较差, 而基于深度学习的点云几何编码算法多数采用自编码器 (AutoEncoder, AE) 神经网络 架构, 存在一定程度的特征信息丢失问题。 此外, 近年研究多数聚焦于熵编码阶段的改进, 却忽视对点云几何空间与 其潜在特征空间转换的优化。 针对以上问题, 提出一种基于可逆神经网络 (Invertible Neural Network, INN) 的点云几何 有损编码算法, 其采用具有数学上严格可逆属性的 INN 进行点云几何信息的特征提取, 避免编码过程中的信息丢失, 保证解码过程中重建点云的稳定性。 设计 3D-Dense-Block 与通道紧缩模块, 用于强化特征信息、增加算法网络的非线 性表达能力, 同时避免训练中次优解的出现。 实验结果表明, 该算法在点云编码公开数据集———微软公司上半身体素 化点云集 (Microsoft Voxelized Upper Bodies, MVUB) 和运动图像专家组织 (Moving Picture Experts Group, MPEG) 8i 数据集 上实现了优于 MPEG 基准算法的率失真性能。 [ABSTRACT FROM AUTHOR] |
| Copyright of Radio Communications Technology is the property of 54th Research Institute of CETC and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Datenbank: | Complementary Index |
Nájsť tento článok vo Web of Science Schreiben Sie den ersten Kommentar!