面向点云无损压缩的快速细节层次优化方法

Optimization Method for Level of Detail of Lossless Point Cloud Compression

为解决点云压缩中基于距离的细节层次(LoD)划分未优化的问题,提出了基于预测残差的LoD优化模型以及相应的快速LoD划分方法。推导了LoD预测残差与码率的关系,建立了LoD预测残差和距离的数学模型;以预测残差的最小化为目标,通过对点云进行预编码或在线计算获取模型参数,根据所提模型获得编码性能最优的LoD层数;为了降低实现复杂度,分析模型各参数对编码性能的影响,得出细节层中的点数随着LoD层数的增加而呈现指数级减少,导致这部分点对编码性能的影响急剧降低这一结论;根据均匀采样及平滑点云的特性,对模型参数进行简化,通过求解细节层中的点数占点云中全部点数的比例得到编码性能最优的LoD层数;在此基础上,提出了一种基于阈值的LoD划分快速方法,提高了模型的实用性。实验结果表明:所提方法在不增加码率的情况下,有效降低了编解码复杂度,平均节省了约4%的编码时间和约6%的解码时间。

To improve the distance-based level of detail(LoD)construction in point cloud compression,a prediction residual-based LoD optimization model and a fast LoD generation method are proposed.The relationship between LoD prediction residuals and coding bitrate is firstly deduced.A model of LoD prediction residual and distance is further constructed,and the model parameters can be obtained by pre-coding or online calculation of the point cloud.Relying on this model,the optimal number of LoD layers can be obtained by minimizing the prediction residuals.To lower the complexity of the proposed method,the influences of the model parameters on the coding performance are analyzed.It is found that the number of points in detail layer decreases exponentially with the increasing LoD layers,which makes the impact on coding performance weaken sharply.Following this analysis,the model parameter determination is simplified according to uniform sampling and smooth distribution features of point clouds.The optimal number of LoD layers can be obtained by the proportion of the points in the detail layer to the total points in the point cloud,so as to achieve the optimal coding performance.Furthermore,a fast LoD generation method based on threshold control is proposed to heighten the model practicability.Experimental results show that the proposed method enables to shorten encoding time by 4%and decoding time by 6%without any performance loss.