基于ONR-CNN的点云属性视频环路滤波算法优化  

Optimization of Point Cloud Attribute Video Loop Filtering Algorithm Based on ONR-CNN

作　　者：张东旭 ZHANG Dongxu(North China University of Technology,Beijing 100144,China)

机构地区：[1]北方工业大学,北京100144

出　　处：《北京工业职业技术学院学报》2024年第3期26-31,共6页Journal of Beijing Polytechnic College

摘　　要：通过对基于视频的点云压缩(V-PCC)投影生成的属性视频进行研究,提出了基于ONR-CNN的点云视频环路滤波算法。算法引入占用信息优化损失函数和加强占用区域编码单元的色度信息表达,提高深度学习网络对占用区域的关注度;根据V-PCC在AI配置下的编码结构,引入迭代训练机制,在训练过程中考虑该编码结构下P帧对I帧的依赖关系,使训练出的网络模型更好地适应编码需求。实验结果表明:在AI配置下,与V-PCC参考软件相比,所提算法在Y,U,V下的BD-AttrRate分别降低5.9%、24.5%和23.5%,BD-TotalRate分别降低4.7%、20.6%和19.0%。By studying the attribute video generated by the projection of Video-based Point Cloud Compression(V-PCC),a point cloud video loop filtering algorithm based on ONR-CNN is proposed.The algorithm introduces occupancy information to optimize the loss function and enhance the chroma information expression of the coding unit in the occupied area,improving the attention of the deep learning network to the occupied area.According to the coding structure of V-PCC in the AI configuration,an iterative training mechanism is introduced.During the training process,the dependency relationship of P-frames on I-frames under this coding structure is considered,so that the trained network model can better adapt to the coding requirements.Experimental results show that under the AI configuration,compared with the V-PCC reference software,the BD-AttrRate of the proposed algorithm in Y,U,and V is reduced by 5.9%,24.5%,and 23.5%respectively,and the BD-TotalRate is reduced by 4.7%,20.6%,and 19.0%respectively.

关 键 词：基于视频的点云压缩 点云属性视频 深度学习 

分 类 号：TN919.81[电子电信—通信与信息系统]