Research On High Efficiency Compression Algorithms For 3D Point Cloud

In recent years,with the development of multimedia communication and 3D imaging technologies,3D point cloud has been widely applied in many industries,such as 3D reconstruction,simulation,augment reality,immersion communication,and so on.3D point cloud is composed of huge amount of point sets presenting the detail 3D positions,where each point has a/multiple feature?e.g.,color,normal,etc.?.As a new space data,3D data is very suitable for presenting 3D model and space with high efficiency.Although 3D point cloud has such good merits,it has huge amount of data.How to develop high efficiency 3D point compression schemes for its storage and transmission has been an emerging topic.Therefore,the research on high efficiency compression algorithms for 3D point cloud has important theoretical and practical value.This thesis studies the high efficiency compression algorithms for 3D point cloud based on its characteristics,including one-dimensional?1D?DCT compression algorithm based on Morton sorting,3D point cloud compression method using improved graph transform,geometric guided sparse representation 3D Point cloud compression algorithm.Their details can be described as follows:1.Considering that the scattered point clouds can be arranged into 1D point array by reordering,a 1D-DCT compression algorithm based on Morton ordering is proposed.The proposed method firstly divides the original input point cloud by the octree,and treats the blocks as units.Since the space distribution of 3D point cloud is irregular,the Mordon order?also known as Z-scan?is used to rearrange irregular points so as to preserve the correlation between adjacent points as much as possible.Then,the 1D-DCT with better de-correlation performance is exploited to transform the aligned 1D color signals.In the entropy coding process,the DC coefficients with larger amplitudes are encoded by the predictive coding,and the remaining AC coefficients with smaller amplitudes are quantized directly.Finally,the prediction residual and the quantized AC coefficient are arithmetically coded.Experimental results show that the proposed method can effectively compress the 3D point cloud data.2.Considering that the scattered points can be connected into graphs by adding edges,a 3D point cloud compression method using improved graph transform is proposed.This method first uses KD-Tree to segment the original input point cloud,and the construction of the graphics is then optimized so that each block produces only a unique graph.After the graph is generated,the graph transform matrix is calculated and the de-averaged color signal is transformed.In the entropy coding process,a block mean prediction method is designed,including five angle modes and one DC mode.The best prediction mode and prediction residual are obtained by comparing the prediction residuals of the respective reference blocks.Since the proportion of the zero coefficient of the quantized transform coefficient is relatively large,the Run-Level method is employed.Finally,the best prediction mode,the quantized prediction residual,and the Run-Level coding parameters are all encoded by arithmetic coding.Experimental results show that the proposed algorithm has higher compression efficiency than multiple existing 3D point cloud compression algorithms.3.Considering that the sparse representation can effectively represent and compress high-dimensional signals,a geometric guided sparse representation 3D point cloud compression algorithm is proposed.This method follows the segmentation of octrees and treats the irregular distribution of points within each block as an adaptive sampling process.Then,the geometric guided sparse representation is used to transform the point cloud compression problem into the ₀l norm optimization problem,which is solved by the OMP algorithm.In the entropy coding process,a block mean prediction method is designed,which includes 8 angle modes and a DC mode.The best prediction mode and prediction residual are obtained by comparing the prediction residuals of the respective reference blocks.Since the solved coefficients are sparse,the quantized coefficients are still encoded using the Run-Level encoding scheme.Finally,the best prediction mode,the quantized prediction residual,and the Run-Level coding parameters are all encoded by arithmetic coding.Experimental results shows the superiority of the proposed algorithm on compression of 3D point cloud.In summary,this thesis mainly explores the 3D point cloud high efficiency compression algorithms based on its characteristics,which contains novelty and challenge to some extent.The research results of this thesis have opened up a vision for the study on the compression technology of 3D point cloud to a certain extent,having important theoretical research significance and practical application value.