TY - JOUR
T1 - 面向点云无损压缩的快速细节层次优化方法
AU - Wei, Lei
AU - Wan, Shuai
AU - Wang, Zhecheng
AU - Ding, Xiaobin
AU - Zhang, Wei
N1 - Publisher Copyright:
© 2021, Editorial Office of Journal of Xi'an Jiaotong University. All right reserved.
PY - 2021/9/10
Y1 - 2021/9/10
N2 - 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.
AB - 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.
KW - Coding performance
KW - Level of detail
KW - Lossless compression
KW - Point cloud compression
KW - Prediction residuals
UR - http://www.scopus.com/inward/record.url?scp=85114838072&partnerID=8YFLogxK
U2 - 10.7652/xjtuxb202109010
DO - 10.7652/xjtuxb202109010
M3 - 文章
AN - SCOPUS:85114838072
SN - 0253-987X
VL - 55
SP - 88
EP - 96
JO - Hsi-An Chiao Tung Ta Hsueh/Journal of Xi'an Jiaotong University
JF - Hsi-An Chiao Tung Ta Hsueh/Journal of Xi'an Jiaotong University
IS - 9
ER -