点云
代表(政治)
计算机科学
人工智能
特征(语言学)
点(几何)
降噪
失真(音乐)
利用
接头(建筑物)
模式识别(心理学)
功能(生物学)
计算机视觉
算法
噪音(视频)
钥匙(锁)
数学
云计算
坐标系
还原(数学)
计算
图像去噪
特征提取
权函数
噪声数据
数据挖掘
出处
期刊:
[SPIE]
日期:2026-10-06
卷期号:: 131-131
摘要
Most existing point cloud denoising methods only utilize geometric coordinate information and fail to fully exploit auxiliary information such as color, which easily leads to over-smoothing or structural distortion in complex scenes. This paper proposes a Color-Guided Neighborhood Reweighting module (CGNR) that considers both geometric and color similarities in neighborhood feature extraction. The module adaptively weights neighborhood features through a joint weight function to enhance representation capability in complex structural regions. The CGNR module adopts a non-parametric design without additional learnable parameters and is easy to integrate. Experimental results on the ShapeNetSem, Greyc, and RueMadame datasets demonstrate that CGNR effectively improves denoising performance and structure preservation ability with nearly no increase in computational overhead.
科研通智能强力驱动
Strongly Powered by AbleSci AI