WO2019153342A1 - 一种基于增强图变换的点云属性压缩方法 - Google Patents
一种基于增强图变换的点云属性压缩方法 Download PDFInfo
- Publication number
- WO2019153342A1 WO2019153342A1 PCT/CN2018/076495 CN2018076495W WO2019153342A1 WO 2019153342 A1 WO2019153342 A1 WO 2019153342A1 CN 2018076495 W CN2018076495 W CN 2018076495W WO 2019153342 A1 WO2019153342 A1 WO 2019153342A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- point cloud
- transformation
- graph
- points
- matrix
- Prior art date
Links
- 230000009466 transformation Effects 0.000 title claims abstract description 113
- 238000007906 compression Methods 0.000 title claims abstract description 53
- 230000006835 compression Effects 0.000 title claims abstract description 52
- 238000000034 method Methods 0.000 title claims abstract description 48
- 230000003595 spectral effect Effects 0.000 claims abstract description 12
- 238000010183 spectrum analysis Methods 0.000 claims abstract description 8
- 238000003672 processing method Methods 0.000 claims abstract description 3
- 238000013139 quantization Methods 0.000 claims abstract description 3
- 239000011159 matrix material Substances 0.000 claims description 58
- 238000006243 chemical reaction Methods 0.000 claims description 14
- 238000000354 decomposition reaction Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 238000000638 solvent extraction Methods 0.000 claims description 6
- 238000001228 spectrum Methods 0.000 claims description 6
- 230000001131 transforming effect Effects 0.000 claims description 5
- 238000005192 partition Methods 0.000 claims description 2
- 230000000007 visual effect Effects 0.000 claims description 2
- 230000002708 enhancing effect Effects 0.000 claims 1
- 238000012805 post-processing Methods 0.000 claims 1
- 230000011218 segmentation Effects 0.000 claims 1
- 238000012360 testing method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000011426 transformation method Methods 0.000 description 2
- WYTGDNHDOZPMIW-RCBQFDQVSA-N alstonine Natural products C1=CC2=C3C=CC=CC3=NC2=C2N1C[C@H]1[C@H](C)OC=C(C(=O)OC)[C@H]1C2 WYTGDNHDOZPMIW-RCBQFDQVSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 208000018375 cerebral sinovenous thrombosis Diseases 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013144 data compression Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000000059 patterning Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
- G06T9/007—Transform coding, e.g. discrete cosine transform
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/169—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding
- H04N19/186—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being a colour or a chrominance component
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2323—Non-hierarchical techniques based on graph theory, e.g. minimum spanning trees [MST] or graph cuts
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/90—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using coding techniques not provided for in groups H04N19/10-H04N19/85, e.g. fractals
- H04N19/96—Tree coding, e.g. quad-tree coding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
- H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
- H03M7/40—Conversion to or from variable length codes, e.g. Shannon-Fano code, Huffman code, Morse code
- H03M7/4031—Fixed length to variable length coding
- H03M7/4037—Prefix coding
- H03M7/4043—Adaptive prefix coding
- H03M7/405—Tree adaptation
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
- H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
- H03M7/70—Type of the data to be coded, other than image and sound
Definitions
- the invention belongs to the technical field of point cloud data processing, and relates to a point cloud data compression method, in particular to a method for point cloud attribute compression based on enhanced graph transformation.
- a three-dimensional point cloud is an effective form of expressing the three-dimensional structure of the real world. With the rapid development of three-dimensional scanning devices (laser, radar, etc.), the accuracy and resolution of the point cloud are higher, making it possible to digitize the three-dimensional information of the real world with high efficiency and high precision.
- High-precision point clouds are widely used in many popular fields such as smart cities, unmanned driving, and cultural relics protection.
- the point cloud is obtained by sampling the surface of the object by the 3D scanning device.
- the number of points of a frame point cloud is generally one million, and each point contains geometric information and attribute information such as color and texture, and the amount of data is very large. The huge amount of data in the 3D point cloud brings great challenges to data storage and transmission, so point cloud compression is necessary.
- Point cloud compression is mainly divided into geometric compression and attribute compression.
- the existing point cloud attribute compression framework mainly includes the following:
- This method first uses the octree to spatially decompose the point cloud to obtain the coded block, then performs depth-first traversal on the tree, and writes the traversed node color values in a serpentine order.
- Into the two-dimensional JPEG table and then use the existing JPEG encoder to encode the obtained point cloud color table, wherein the JPEG encoder uses DCT.
- the method utilizes the existing encoder, and the computational complexity is low, but the spatial correlation between points is not fully utilized, and the compression performance needs to be improved;
- the method first uses the octree to spatially decompose the point cloud, and divides it into a specific level to obtain a transform block; forming a graph in each transform block, along any axis The two points whose distance is less than 1 are connected by one edge, and the weight of the edge is inversely proportional to the Euclidean distance; then the attribute information of the node in the figure is transformed.
- the compression performance of the method is good; but the computational complexity is high, and the patterning method may bring subgraph problems, affecting the efficiency of graph transformation, and there is still room for improvement;
- This method first uses the KD tree to spatially decompose the point cloud, and divides it to a certain depth to obtain a transform block, wherein each transform block contains approximately the same number of points: within the transform block Each two points are connected by one edge, and the weight of the edge is related to the Euclidean distance.
- the set edge distance Euclidean distance threshold determines the sparseness of the graph; then the graph transformation of the attribute information of the point in the graph is performed.
- the method completely solves the subgraph problem, and at the same time, the compression performance is improved compared with the former two methods, but the operation complexity is high and the performance still needs to be improved.
- the present invention provides a method for point cloud attribute compression based on enhanced graph transformation, first using a K-dimension (KD) tree to perform spatial division of a point cloud. Then, using the graph transform processing based on spectrum analysis, the graphs in the coding block are clustered, and the local graph transform scheme is extended based on the existing graph transform to form an enhanced graph transform with two transform modes, and the graph transform is improved. Compression performance.
- KD K-dimension
- a point cloud attribute compression method based on enhanced graph transformation For point cloud data, a KD tree is first used to divide the point cloud spatial domain to obtain a coding block, and a new graph transform processing method combined with spectrum analysis is proposed. On the map inside, the point cloud is further clustered, and the local graph transformation scheme is extended based on the existing graph transformation to form an enhanced graph transformation with two transformation modes. The optimal mode is selected by the mode decision to reach the attribute.
- the best performance for compression includes the following steps:
- Color space conversion of point cloud attribute read the point cloud attribute information to be processed, consider the visual characteristics of the human eye and the difficulty of compression processing, and convert the point cloud color space from RGB space to YUV space;
- KD tree division is performed on the point cloud according to the geometric information. Each time the coordinate axis with the largest variance of the point cloud position coordinates is selected as the division axis, and the point where the coordinate size is the median value is selected. As a dividing point, iteratively divides until the set KD tree depth is reached; the block obtained by the last layer of the KD tree partition is the coded block of the point cloud, and the coded blocks are numbered according to the order of the breadth traversal;
- Enhanced graph transformation based on spectrum clustering transforming the coded blocks according to the numbering sequence of the coding blocks, and performing spectral map clustering on the basis of the existing graph transform to form an enhanced graph transform, and providing two transform modes in total;
- every two points n i , n j are connected by an edge ⁇ ij to construct a full-point graph G, wherein the weight of the edge is determined by the geometric position of the two points; the weight of the edge ⁇ ij is ⁇ ij reflected between n i, n j of the two geometrical correlation, weighting matrix W ⁇ configuration adjacent the graph, further eigenvector matrix a;
- Transform mode 2 The eigenvector matrix A of graph G reflects the spectral distribution of the graph; the sign of the second dimension vector of matrix A is used to cluster all the points in the block into two categories, such as the number of two types of points.
- the graph G is divided into two partial maps G 1 and G 2 according to the clustering condition, and the corresponding eigenvector matrices A 1 and A 2 are respectively obtained; 1.
- the color information of the points in G 2 is transformed using the corresponding transformation matrix;
- Transform mode decision There are two modes for transforming the color information of the coded block. It is necessary to estimate the performance of the transform to make the mode decision, select the best transform mode, and calculate the absolute value of the first k largest coefficients in the transformed coefficient. And the proportion of the sum of the absolute values of the transform coefficients as the fraction of the transform mode; the higher the score, the higher the proportion of the selected transform coefficients in the sum of the transform coefficients, the higher the transform efficiency representing the mode, the performance The better, the mode with the largest score will be selected as the transform mode of the current block;
- the point p i in the point cloud has the color values of the RGB color space r i , g i , b i , and the RGB is converted into the YUV color space by the color space conversion matrix, and the color values are y i , u i , v i ;
- the specific process of dividing and numbering the coding blocks in the above step 2) is as follows: when the KD tree is divided, the coordinate axis with the largest distribution variance of the selected points is used as the division axis, and the correlation of each point in the transformation block is fully considered; The point on which the coordinate on the axis is the median value is used as the division point, so that the number of points in the transformation block is substantially the same.
- the point cloud to be processed has a total of N points, and the KD tree has a division depth of d. After dividing the point cloud by d times, 2d coding blocks are obtained; all coding blocks are numbered according to the order of breadth traversal.
- Equation 2 (-1) Construct a graph G in each transform block, with each edge n i , n j connected by an edge ⁇ ij , the weight of the edge ⁇ ij ⁇ ij and the Euclidean between the two points Distance related, often expressed as Equation 2:
- the parameter ⁇ reflects the variance of the current point cloud distribution
- the parameter T is a distance threshold for determining whether the line is connected between the two points, determines the sparsity of the Tula's matrix, and generally uses an empirical set value
- the neighboring matrix W of the graph G is a set of edge weights ⁇ ij , reflecting the correlation between points in the transform block;
- the density matrix D of the graph G is a diagonal matrix, where D i is the neighboring matrix
- Equation 3 The sum of non-zero elements in the i-line is expressed as Equation 3, reflecting the density of the correlation between the i-th point and other points;
- the transform operator of Figure G uses the Laplacian matrix L, expressed as Equation 4:
- Equation 5 (-3) Perform eigen decomposition on the Laplacian matrix L to obtain the eigenvector matrix A.
- the global graph transformation matrix as the transformation mode one is used to compress the attribute information of the point cloud, which is expressed as Equation 5:
- Equation 6 the global map G is divided into two partial maps G 1 and G 2 , which are composed of corresponding points n and edges ⁇ , which are expressed as Equation 7:
- count(C 1 ) is the number of points in C 1
- count(C 2 ) is the number of points in C 2
- count(block) is the total number of points in the current coded block.
- the specific process of the transformation mode decision described in the above step 4) is as follows: first, the absolute value of the transformed coefficient Trans in step 3) is arranged in descending order, and then the sum of the absolute values of the maximum k-dimensional maximum coefficients is calculated according to Equation 8 in the transform coefficient.
- the performance of point cloud attribute compression is measured by the code rate and Peak Signal to Noise Ratio (PSNR), where the code rate is obtained by dividing the total number of bits of the code word by the number of points of the point cloud.
- PSNR Peak Signal to Noise Ratio
- Bpp bits per point
- the unit of PSNR is decibel dB; the smaller the code rate, the larger the PSNR, and the better the compression performance of the point cloud attribute.
- the above method first uses the KD tree to divide the point cloud into the airspace, and then uses the spectrum clustering of the graph to perform frequency domain partitioning, and uses the enhanced graph transformation scheme to compress the point cloud attribute, providing two transformation modes, and selecting the best by the mode decision. Mode to achieve better point cloud compression performance.
- the invention provides a point cloud attribute compression method based on enhanced graph transformation, which has the following technical advantages:
- the point cloud is first used to divide the airspace by using the KD tree. Based on the spectrum analysis of the graph, the coded block is frequency-domain partitioned to achieve more accurate point cloud partitioning, which provides guarantee for more point cloud compression performance.
- FIG. 1 is a flow chart of a method provided by the present invention.
- FIG. 2 is a flow chart showing an example of a point cloud coded block enhancement map transform process.
- FIG. 3 is a diagram showing an example of a code stream structure after compression of point cloud attribute information.
- Figures 4a, b and c are graphs comparing the compression performance of the method of the invention with prior art methods.
- Figure 4a is a comparison of the compression performance of the test longdress_vox10_1300.ply;
- Figure 4b is a comparison of the compression performance of the test Queen_frame_0200.ply;
- Figure 4c is a comparison of the compression performance of the test soldier_vox10_0690.ply.
- a method for point cloud attribute compression based on enhanced graph transformation uses a K-dimension (KD) tree to perform spatial domain partitioning on a point cloud first, and proposes a new graph transformation processing combined with spectrum analysis.
- the method performs spectral clustering on the point cloud on the map in the point cloud coding block, expands the local map transformation scheme based on the existing graph transformation, forms an enhanced graph transform with two transform modes, and improves the compression of the graph transform.
- Figure 1 is a block flow diagram of the method of the present invention.
- the point cloud attribute compression is performed by the method of the present invention, as shown in FIG. 1 , the specific implementation steps are:
- Color space conversion of point cloud attribute read the point cloud attribute information to be processed, the point p i in the point cloud has the color values of RGB color space r i , g i , b i , through the color space conversion matrix Convert RGB to YUV color space, the color values are y i , u i , v i , as shown in Equation 1:
- the RGB color value of the first point p 1 of the point cloud longdress_vox10_1300.ply is (102, 94, 87), and the YUV color value is obtained by the color conversion matrix (54.4128, -2.7926, 50.3798).
- the RGB color value of the first point p 1 of the point cloud Queen_frame_0200.ply is (102, 80, 71), and the YUV color value is obtained by the color conversion matrix (48.0172, 9.8702, 44.1126).
- the RGB color value of the first point p 1 of the point cloud soldier_vox10_0690.ply is (68, 65, 64), and the YUV color value is obtained by the color conversion matrix (39.0078, -5.4862, 34.4784).
- the KD tree is essentially a binary tree.
- the coordinate axis with the largest variance of the distribution in the position coordinates of the point cloud is selected as the division axis.
- a point whose coordinate size is a median value is selected as a division point, and iteratively divides until the set KD tree depth is reached, and the completed KD tree and the coded block are numbered.
- the point cloud longdress_vox10_1300.ply has a total of 857966 points, and the KD tree division depth d is set to 13, and the number of points in the block after division is 104 or 105.
- the point cloud Queen_frame_0200.ply has a total of 1000993 points, and the KD tree division depth d is set to 13, and the number of points in the block after division is 122 or 123.
- the point cloud soldier_vox10_0690.ply has a total of 1089091 points, the KD tree division depth d is set to 13, and the number of points within the block after division is 132 or 133.
- Enhanced graph transformation based on spectral clustering The point cloud is divided by the KD tree of step (2) to obtain a coded block, and then processed by using an enhanced graph transform containing two transform modes, and the processing flow is as shown in FIG. 2 .
- the essence of the enhanced graph transform is that the spectrum clustering of the extended graph is another transform mode based on the current graph transform. The specific details are as follows:
- Adjacent matrix of the graph is a set of edge weights ⁇ ij between two points n i and n j , reflecting the correlation between the points in the transform block, the weight of the edges Determined by Equation 2:
- the parameter ⁇ is the variance of the geometrical coordinates of the point cloud
- the parameter T is a distance threshold for determining the correlation between the two points, affecting the generation of the transformation matrix, generally used It means that T, ⁇ ' is set to 0.8.
- Density matrix of graph The density matrix D of graph G is a diagonal matrix, and the expression is as shown in Equation 3, where D i is the sum of non-zero elements in the i-th row of the adjacent matrix, reflecting the The density of the correlation of i points with other points:
- Equation 6 Spectral clustering based on graph transformation matrix: spectral clustering is performed on the second dimension vector V 2 of the feature vector matrix A, and the points are clustered into two types C 1 according to the sign of the vector parameter value p i C 2 is expressed as Equation 6; according to the clustering situation, the global map G is divided into two partial maps G 1 and G 2 , which are composed of corresponding points n and edges ⁇ , which are expressed as Equation 7:
- count(C 1 ) is the number of points in C 1
- count(C 2 ) is the number of points in C 2
- count(block) is the total number of points in the current coded block.
- the graph transformation matrix of the transformation mode 2 eigen-decomposition of the two partial maps G 1 and G 2 respectively, and obtaining the transformation matrices A 1 and A 2 as the partial graph transformation matrix of the transformation mode 2, for corresponding The transformation of the color information of the points in the figure.
- the first coding block b 1 of the point cloud longdress_vox10_1300 uses the fractions J of the transformation modes one and two to be 0.58 and 0.4, respectively, so J is selected to be a larger mode 1 as the optimal transformation mode of the b 1 block.
- the method of the present invention is significantly superior to the existing mainstream methods in attribute compression performance on three typical types of point cloud sequences tested (based on octree and DCT attribute compression) , RN Mekuria, K. Blom, and P. Cesar, "Design, Implementation and Eva] uation of a Point Cloud Codec for Tele-Immersive Video.” IEEE Trans.CSVT, vol. PP, no. 99, pp.1- 1, 2016.), PSNR improved by 1-4 dB under the same bit rate.
- the calculation method is slightly larger than the current method, the compression performance has obvious advantages, and the subgraph problem in the previous graph transformation is overcome, and the advantages are outstanding.
- the invention provides a method for point cloud attribute compression based on enhanced graph transformation, which first uses a K-dimension (KD) tree to perform spatial domain partitioning on a point cloud, and then uses a graph transform processing based on spectrum analysis to perform spectrum on a graph in a coded block.
- KD K-dimension
- Clustering based on the existing graph transformation, expands the implementation of the local graph transformation scheme, forms an enhanced graph transform with two transform modes, and improves the compression performance of the graph transform.
- three-dimensional scanning devices laser, radar, etc.
- the accuracy and resolution of the point cloud are higher, making it possible to digitize the three-dimensional information of the real world with high efficiency and high precision.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Signal Processing (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Discrete Mathematics (AREA)
- Computing Systems (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Geometry (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
- Image Processing (AREA)
- Compression Of Band Width Or Redundancy In Fax (AREA)
Abstract
一种基于增强图变换的点云属性压缩方法,针对点云属性信息,使用K-dimension(KD)树对点云先进行空域划分,提出一种新的结合频谱分析的图变换处理方法,在点云编码块内的图上对点云再进行频谱聚类,在现有图变换基础上拓展实现局部图变换方案,形成具有两种变换模式的增强图变换,提高图变换的压缩性能;包括:点云属性的颜色空间转换;采用K-dimension(KD)树划分点云得到编码块;基于频谱聚类的增强图变换;变换模式决策;均匀量化和熵编码。提出一种新的基于频谱分析的增强图变换方案,其中包括两种变换模式,由模式决策选出最佳模式;对点云进行树划分后,在各个编码块内均构建一张图并使用图变换作为变换模式一,在此基础上实现图频聚类,把图划分成两张局部图再进行局部图变换作为变换模式二,支持两种变换模式的增强图变换方案由模式决策选择出最佳模式,以达到点云属性压缩的最佳性能。
Description
本发明属于点云数据处理技术领域,涉及点云数据压缩方法,尤其涉及一种基于增强图变换的点云属性压缩的方法。
三维点云是一种表达现实世界三维结构的有效形式。随着三维扫描设备(激光、雷达等)的快速发展,点云的精度、分辨率更高,使得高效率、高精度地将现实世界的三维信息数字化变成可能。高精度点云广泛应用于智慧城市、无人驾驶、文物保护等众多热门领域。点云是三维扫描设备对物体表面采样所获取的,一帧点云的点数一般是百万级别,其中每个点包含几何信息和颜色、纹理等属性信息,数据量十分庞大。三维点云庞大的数据量给数据存储、传输等带来巨大挑战,所以点云压缩十分必要。
点云压缩主要分为几何压缩和属性压缩,现有的点云属性压缩框架主要包括以下几种:
一、基于八叉树分解和DCT的方法:该方法先用八叉树对点云进行空间分解得到编码块,然后对树进行深度优先遍历,将遍历的结点颜色值按照蛇形的顺序写入二维的JPEG表,再利用现有的JPEG编码器对获得的点云颜色表进行编码,其中JPEG编码器使用的是DCT。该方法利用现有的编码器,计算复杂度低,但并未充分利用点与点之间的空间相关性,在压缩性能上有待提高;
二、基于八叉树分解和图变换的方法:该方法先用八叉树对点云进行空间分解,划分到特定层次得到变换块;在每个变换块内形成图,将沿着任意坐标轴的距离不超过1的两点用一条边连接起来,边的权重与欧氏距离成反比;再对图中结点的属性信息进行图变换。该方法的压缩性能佳;但运算复杂度较高,其构图方式可能会带来子图问题,影响图变换的效率,仍有改善的空间;
三、基于KD树分解和图变换的方法:该方法先用KD树对点云进行空间分解,划分到一定深度得到变换块,其中每个变换块内所包含的点数大致相同:在变换块内,每两点用一条边连接起来,边的权重与欧氏距离有关,设定的边欧式距离阈值决定图的稀疏度;然后再对图内点的属性信息进行图变换。该方法彻底解决了子图问题,同时在压缩性能方面较前两种方法都有较大的提升,但运算复杂度较高,性能仍待改善。
发明的公开
为了克服上述现有技术的不足,进一步改善点云属性的压缩性能,本发明提供一种基于增强图变换的点云属性压缩的方法,先使用K-dimension(KD)树对点云进行空域划分,后使用基于频谱分析的图变换处理对编码块内的图进行频谱聚类,在现有图变换基础上拓展实现局部图变换方案,形成具有两种变换模式的增强图变换,提高图变换的压缩性能。
本发明提供的技术方案是:
一种基于增强图变换的点云属性压缩方法,针对点云数据,先使用KD树对点云空域划分得到编码块,提出一种新的结合频谱分析的图变换处理方法,在点云编码块内的图上对点云再进行频谱聚类,在现有图变换基础上拓展实现局部图变换方案,形成具有两种变换模式的增强图变换,由模式决策选择出最佳模式,以达到属性压缩的最佳性能;包括如下步骤:
1)点云属性的颜色空间转换:读入待处理的点云属性信息,考虑人眼的视觉特性和压缩处理的难易程度,将点云颜色空间从RGB空间转换到YUV空间;
2)采用KD树划分点云得到编码块:根据几何信息对点云进行KD树划分,每次选择点云位置坐标中分布方差最大的坐标轴作为划分轴,选取坐标大小是中位值的点作为划分点,迭代划分直至达到设定的KD树深度;KD树划分的最后一层所得到的块即为点云的编码块,按照广度遍历的顺序对编码块进行编号;
3)基于频谱聚类的增强图变换:根据编码块的编号顺序依次对编码块进行变换处理,在现有图变换基础上扩展图的频谱聚类构成增强图变换,共提供两种变换模式;在编码块内,每两个点n
i、n
j之间用边ε
ij连接起来构建一张全点图G,其中边的权重由两点的几何位置决定;边ε
ij的权重大小ω
ij反映n
i、n
j两点之间的几何相关性,权重ω构成图的邻近矩阵W,进一步得到特征向量矩阵A;
变换模式一:将该特征向量A作为图的变换矩阵,对编码块的颜色信息进行变换;
变换模式二:图G的特征向量矩阵A,反映的是图的频谱分布;使用矩阵A的第二维列向量的正负号将块内所有点聚成两类,如两类点的数量均达到块内点总数的40%以上,则根据聚类情况将图G划分成两张局部图G
1、G
2,再分别得到对应的特征向量矩阵A
1、A
2;对两张局部图G
1、G
2中的点的颜色信息,使用对应的变换矩阵进行变换;
4)变换模式决策:对编码块的颜色信息进行变换有两种模式,需要估计变换的性能来进行模式决策,选出最佳的变换模式;计算变换后系数中前k个最大系数绝对值之和在变换系数绝对值总和中所占的比例,作为变换模式的分数;分数越高,代表选定的变换系数在变换系数总和中所占比例越高,代表该模式的变 换效率越高,性能越好,具有最大分数的模式将被选择为当前块的变换模式;
5)点云属性压缩码流的生成:按照顺序迭代处理所有编码块,对变换后系数进行量化,结合变换模式信息进行熵编码,得到点云属性压缩的最终码流;
上述步骤1)所述的颜色空间转换,其具体过程如下:
(1-1)点云中的点p
i具有RGB颜色空间的颜色值为r
i、g
i、b
i,通过颜色空间转换矩阵将RGB转换到YUV颜色空间,颜色值为y
i、u
i、v
i;
(1-2)颜色空间转换的数学表达式为:
上述步骤2)所述的编码块的划分和编号,其具体过程如下:KD树划分时,选择点的分布方差最大的坐标轴作为划分轴,充分考虑了变换块内各点的相关性;同时把划分轴上坐标是中位值的点作为划分点,使变换块内点的数量基本相同。设待处理的点云共有N个点,KD树设定的划分深度为d,经过对点云d次划分后,得到2
d个编码块;对所有的编码块按照广度遍历的顺序进行编号
上述步骤3)中所述的增强图变换,其具体过程如下:
(3-1)在每个变换块内构建一张图G,每两点n
i、n
j之间用一条边ε
ij连接,边ε
ij的权重大小ω
ij与两点之间的欧氏距离相关,常表现为式2:
其中,参数σ反映了当前点云分布的方差;参数T是判定两点之间是否连线的距离阈值,决定图拉普拉斯矩阵的稀疏度,一般使用经验设定值;
(3-2)图G的邻近矩阵W是边权重ω
ij的集合,反映变换块内各点之间的相关性;图G的密度矩阵D是一个对角矩阵,其中D
i是邻近矩阵第i行中非零元素的和,表示为式3,反映了第i个点与其他点相关性的密度;图G的变换算子采用拉普拉斯矩阵L,表示为式4:
D
i=∑
jω
i,j (式3)
L=D-W (式4)
(3-3)对拉普拉斯矩阵L进行特征分解,得到特征向量矩阵A,作为变换模式一的全局图变换矩阵,用于对点云的属性信息的压缩,表示为式5:
L=AΛA
-1 (式5)
(3-4)对特征向量矩阵A的第二维列向量V
2进行频谱聚类,根据向量的参数值p
i的正负号将点聚成两类C
1、C
2,表示为式6;根据聚类情况将全局图G划分成两张局部图G
1、G
2,由对应的点n和边ε构成,表示为式7:
其中式6中,count(C
1)是计算C
1中的点的数量,count(C
2)是计算C
2中的点的数量,count(block)是计算当前编码块内点的总数。
(3-5)对两张局部图G
1、G
2分别进行特征分解,得到变换矩阵A
1、A
2,作为变换模式二的局部图变换矩阵,用于对应图中点的颜色信息的变换;
(3-6)两种变换模式的图变换构成了增强图变换方案;
上述步骤4)中所述的变换模式决策,其具体过程如下:首先对步骤3)变换后系数Trans的绝对值进行降序排列,然后根据式8计算前k维最大系数绝对值之和在变换系数绝对值总和中所占比例,其中k一般使用经验设定值;比例值J作为变换模式的分数,选择J最高的模式作为最佳变换模式:
上述步骤5)中所述的点云属性压缩码流的生成,其具体过程如下:
(5-1)经过上述步骤1)至4)的处理,再对变换后的系数量化,结合变换模式信息进行熵编码,生成最终的点云属性码流;
(5-2)点云属性压缩的性能由码率和峰值信噪比PSNR(Peak Signal to Noise Ratio)来衡量,其中码率由码字总比特数除以点云的点数可得,单位是bpp(bits per point),PSNR的单位是分贝dB;码率越小,PSNR越大,点云属性压缩性能越好。
上述方法先使用KD树对点云进行空域划分,再使用图的频谱聚类进行频域划分,使用增强图变换方案对点云属性进行压缩,提供两种变换模式,由模式决策选出最佳模式,实现更优的点云压缩性能。
与现有技术相比,本发明的有益效果是:
本发明提供一种基于增强图变换的点云属性压缩方法,具有以下技术优势:
(一)对点云先使用KD树进行空域划分,在此基础上结合图的频谱分析对编码块进行频域划分,实现更精准的点云划分,为实现更加的点云压缩性能提供保障。
(二)在原有图变换的基础上拓展了频谱聚类,新增局部图变换这一变换模式,构成包含两种变换模式的增强图变换,通过模式决策选出最佳变换模式,提高了图变换的效率,改善变换性能。
附图的简要说明
图1是本发明提供方法的流程框图。
图2是对点云编码块增强图变换处理的示例流程图。
其中,(a)点云示例;(b)经KD树划分后得到编码块;(c)编码块示例;(d)对当前编码块进行增强图变换:若满足聚类条件,则将全点图分割成两张局部图;否则,不分割。
图3是点云属性信息压缩后的码流结构示例图。
图4a、b和c是本发明方法与现有传统方法的压缩性能对比图。
其中:
图4a是测试longdress_vox10_1300.ply的压缩性能对比图;
图4b是测试Queen_frame_0200.ply的压缩性能对比图;
图4c是测试soldier_vox10_0690.ply的压缩性能对比图。
实施本发明的最佳方式
下面结合附图,通过实施例进一步描述本发明,但不以任何方式限制本发明的范围。
本发明的一种基于增强图变换的点云属性压缩的方法,针对点云数据,使用K-dimension(KD)树对点云先进行空域划分,提出一种新的结合频谱分析的图变换处理方法,在点云编码块内的图上对点云再进行频谱聚类,在现有图变换基础上拓展实现局部图变换方案,形成具有两种变换模式的增强图变换,提高图变换的压缩性能;图1是本发明方法的流程框图。
以下针对MPEG点云压缩工作组中官方点云数据集longdress_vox10_1300.ply、Queen_frame_0200.ply和soldier_vox10_0690.ply,采用本发明方法进行点云属性压缩,如图1所示,具体实施步骤为:
(1)点云属性的颜色空间转换:读入待处理的点云属性信息,点云中的点p
i具有RGB颜色空间的颜色值为r
i、g
i、b
i,通过颜色空间转换矩阵将RGB转换到YUV颜色空间,颜色值为y
i、u
i、v
i,如式1所示:
点云longdress_vox10_1300.ply的第一个点p
1的RGB颜色值为(102,94,87),经过颜色转换矩阵的处理得到YUV颜色值为(54.4128,-2.7926,50.3798)。
点云Queen_frame_0200.ply的第一个点p
1的RGB颜色值为(102,80,71), 经过颜色转换矩阵的处理得到YUV颜色值为(48.0172,9.8702,44.1126)。
点云soldier_vox10_0690.ply的第一个点p
1的RGB颜色值为(68,65,64),经过颜色转换矩阵的处理得到YUV颜色值为(39.0078,-5.4862,34.4784)。
(2)采用KD树划分点云得到编码块:KD树实质上是一种二叉树,对该点云进行KD树的划分时,每次选择点云位置坐标中分布方差最大的坐标轴作为划分轴,在该轴上选取坐标大小是中位值的点作为划分点,迭代划分直至达到设定的KD树深度,划分完成后的KD树及带编号的编码块。
点云longdress_vox10_1300.ply共有857966个点,KD树划分深度d设为13,经过划分后块内点的数量为104或105。
点云Queen_frame_0200.ply共有1000993个点,KD树划分深度d设为13,经过划分后块内点的数量为122或123。
点云soldier_vox10_0690.ply共有1089091个点,KD树划分深度d设为13,经过划分后块内点的数量为132或133。
(3)基于频谱聚类的增强图变换:点云经步骤(2)的KD树划分后得到编码块,再使用包含两种变换模式的增强图变换进行处理,处理流程如图2所示。增强图变换的实质是在目前图变换的基础上扩展图的频谱聚类为另一种变换模式,具体细节如下:
(3-1)图的构建:在块内每两点之间用边连接起来,构成一张由点n和边ε组成的图G。
(3-2)图的相邻矩阵:图G的相邻矩阵W是n
i、n
j两点之间边权重ω
ij的集合,反映变换块内各点之间的相关性,边的权重由式2确定:
(3-3)图的密度矩阵:图G的密度矩阵D是一个对角矩阵,表达式如式3所示,其中D
i是相邻矩阵第i行中非零元素之和,反映了第i个点与其他点相关性的密度:
D
i=∑
jω
i,j (式3)
(3-4)图的拉普拉斯矩阵:图G的变换算子一般用拉普拉斯矩阵L,其表达式为式4:
L=D-W (式4)
(3-5)变换模式一的图变换矩阵:对拉普拉斯矩阵L进行特征分解,得到特征向量矩阵A作为变换模式一的图变换矩阵,用于对点云的属性信息的压缩,其中特征分解如式5:
L=AΛA
-1 (式5)
(3-6)基于图变换矩阵的频谱聚类:对特征向量矩阵A的第二维列向量V
2进行频谱聚类,根据向量参数值p
i的正负号将点聚成两类C
1、C
2,表示为式6;根据聚类情况将全局图G划分成两张局部图G
1、G
2,由对应的点n和边ε构成,表示为式7:
其中式6中,count(C
1)是计算C
1中的点的数量,count(C
2)是计算C
2中的点的数量,count(block)是计算当前编码块内点的总数。
(3-7)变换模式二的图变换矩阵:对两张局部图G
1、G
2分别进行特征分解,得到变换矩阵A
1、A
2,作为变换模式二的局部图变换矩阵,用于对应图中点的颜色信息的变换。
(3-8)两种变换模式的图变换构成了增强图变换方案。
(4)变换的模式决策:
对编码块的颜色信息有两种变换模式,需要估计变换的性能来进行模式决策,选出最佳的变换模式;计算变换后系数中前k个最大系数绝对值之和在变换系数绝对值总和中所占的比例,作为变换模式的分数,如式8所示;分数越高,代表该模式的变换效率越高,具有最大分数的模式将被选择为当前块的变换模式。
例如,点云longdress_vox10_1300.ply的第一个编码块b
1,使用变换模式一和二的分数J分别是0.58和0.4,所以选择J更大的模式一作为b
1块的最佳变换模式。
(5)点云属性压缩码流的生成:针对点云longdress_vox10_1300.ply的8192个编码块、Queen_frame_0200.ply的8192个编码块、soldier_vox10_0690.ply的8192个编码块,将块内的颜色信息依次经过增强图变换、量化和熵编码处理,再结合变换模式的码流信息,按照编码块的顺序写入码流文件中,最终码流文件的结构如图3所示。点云属性压缩的性能由码率和峰值信噪比PSNR(Peak Signal to Noise Ratio)来衡量,其中码率的单位是bpp(bits per point),PSNR的单位是分贝dB。
为了验证本发明的一种基于增强图变换的点云属性压缩的方法的效果,我们使用上述3个数据集longdress_vox10_1300.ply、Queen_frame_0200.ply、soldier_vox10_0690.ply进行实验,在压缩性能上与现有的方法对比结果如图4a、b和c所示。
从图4a、b和c可以看出,在测试的三类典型的点云序列上,本发明的方法在属性压缩性能上明显优于现有的主流方法(基于八叉树和DCT的属性压缩,R.N.Mekuria,K.Blom,and P.Cesar,“Design,Implementation and Eva]uation of a Point Cloud Codec for Tele-Immersive Video.”IEEE Trans.CSVT,vol.PP,no.99,pp.1-1,2016.),在相同码率的条件下,PSNR改善了1-4dB。本方法虽计算量稍大于目前的方法,但压缩性能的优势明显,而且克服了以往图变换中的子图问题,优点突出。
需要注意的是,公布实施例的目的在于帮助进一步理解本发明,但是本领域的技术人员可以理解:在不脱离本发明及所附权利要求的精神和范围内,各种替换和修改都是可能的。因此,本发明不应局限于实施例所公开的内容,本发明要求保护的范围以权利要求书界定的范围为准。
本发明提供一种基于增强图变换的点云属性压缩的方法,先使用K-dimension(KD)树对点云进行空域划分,后使用基于频谱分析的图变换处理对编码块内的图进行频谱聚类,在现有图变换基础上拓展实现局部图变换方案,形成具有两种变换模式的增强图变换,提高图变换的压缩性能。随着三维扫描设备(激光、雷达等)的快速发展,点云的精度、分辨率更高,使得高效率、高精度地将现实世界的三维信息数字化变成可能。可广泛应用于智慧城市、无人驾驶、文物保护等众多热门领域。
Claims (6)
- 一种基于增强图变换的点云属性压缩方法,针对点云属性信息,使用K-dimension(KD)树对点云先进行空域划分,提出一种新的结合频谱分析的图变换处理方法,在点云编码块内的图上对点云再进行频谱聚类,在现有图变换基础上拓展实现局部图变换方案,形成具有两种变换模式的增强图变换,提高图变换的压缩性能;包括如下步骤:1)点云属性的颜色空间转换:读入待处理的点云属性信息,考虑人眼的视觉特性和压缩处理的难易程度,将点云颜色空间从RGB空间转换到YUV空间;2)采用KD树划分点云得到编码块,并按照广度遍历顺序对编码块进行编号:读入点云的几何信息,根据几何信息对点云进行KD树划分,每次选择点云位置坐标中分布方差最大的坐标轴作为划分轴,选取坐标大小是中位值的点作为划分点,迭代划分直至达到设定的KD树深度;KD树划分的最后一层所得到的块即为点云的编码块,按照广度遍历的顺序对编码块进行编号,该编号将作为编码块后期处理的顺序;3)在编码块内构建图使用增强图变换,有两种变换模式:在编码块内,每两个点n i、n j之间用边ε ij连接起来构建一张全点图G,图上的每个点具有颜色信息,边的权重由两点的几何相对位置决定;边ε ij的权重大小ω ij反映n i、n j两点之间的几何相关性,所有边的权重ω构成图的邻近矩阵W,进一步得到特征向量矩阵A;变换模式一:将全点图的特征向量A作为变换矩阵,对编码块的颜色信息进行变换;变换模式二:图G的特征向量矩阵A反映了图的频谱分布,由此在矩阵A的基础上进行频谱聚类,实现局部图的分割;使用矩阵A的第二维列向量的正负号将块内所有点聚成两类,如两类点的数量均达到块内点总数的40%以上,则根据聚类情况将图G划分成两张局部图G 1、G 2,分别得到对应的特征向量矩阵A 1、A 2;对两张局部图G 1、G 2中的颜色信息,使用对应的变换矩阵进行变换;4)变换的模式决策:对编码块的颜色信息进行变换有两种模式,需要估计变换的性能来进行模式决策,选出最佳的变换模式;计算变换后系数中前k个最大系数绝对值之和在变换系数绝对值总和中所占的比例,作为变换模式的分数;分数越高,代表选定的变换系数在变换系数总和中所占比例越高,代表该模式的变换效率越高,性能越好,具有最大分数的模式将被选择为当前块的变换模式;5)点云属性压缩码流的生成:按照编码顺序处理所有编码块,对变换后系数进行量化,结合变换模式信息进行熵编码,得到点云属性压缩的最终码流;
- 如权利要求1所述点云属性压缩方法,其特征是,步骤3)增强图变换具体过程如下:(4-1)在每个变换块内构建一张图G,每两点n i、n j之间用一条边ε ij连接,边ε ij的权重大小ω ij与两点之间的欧氏距离相关,常用式2计算可得:其中,参数σ反映了当前点云分布的方差;参数τ是判定两点之间是否连线的距离阈值,决定图拉普拉斯矩阵的稀疏度,一般使用经验设定值;(4-2)图G的邻近矩阵W是边权重ω ij的集合,反映块内各点之间的相关性;图G的密度矩阵D是一个对角矩阵,其中D i是邻近矩阵W第i行中非零元素的和,表示为式3,反映了第i个点与其他点相关性的密度;图G的变换算子采用拉普拉斯矩阵L,表示为式4:D i=Σ jω i,j (式3)L=D-W (式4)(4-3)对拉普拉斯矩阵L进行特征分解,表示为式5,得到特征向量矩阵A,作为变换模式一的全局图变换矩阵,用于对点云属性信息的压缩:L=AΛA -1 (式5)其中,A为特征向量矩阵。(4-4)对特征向量矩阵A的第二维列向量V 2进行频谱聚类,根据向量参数值p i的正负号将点聚成两类C 1、C 2,表示为式6;根据聚类情况将全局图G划分成两张局部图G 1、G 2,由对应的点n和边ε构成,表示为式7:其中式6中,count(C 1)是计算C 1中的点的数量,count(C 2)是计算C 2中的点的数量,count(block)是计算当前编码块内点的总数。(4-5)对两张局部图G 1、G 2分别进行特征分解,得到变换矩阵A 1、A 2,作为变换模式二的局部图变换矩阵,用于对应图中点的颜色信息信息的变换;(4-6)两种变换模式的图变换构成了增强图变换方案;
- 如权利要求1所述点云属性压缩方法,其特征是,步骤5)中具体细节如下:(6-1)点云属性信息的码流主要由压缩头信息和编码块信息两大部分组成。其中,头信息主要包括量化步长等;编码块信息流以编码块为单位,按照编码块的顺序排列,每个块内主要包括编码块的变换模式信息和颜色残差信息。(6-2)点云属性压缩的性能由码率和峰值信噪比PSNR(Peak Signal to Noise Ratio)来衡量,其中码率的单位是bpp(bits per point),PSNR的单位是分贝dB;码率越小,PSNR越大,点云属性压缩性能越好。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US16/957,187 US11126887B2 (en) | 2018-02-11 | 2018-02-12 | Enhanced graph transformation-based point cloud attribute compression method |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810142763.8A CN108335335B (zh) | 2018-02-11 | 2018-02-11 | 一种基于增强图变换的点云属性压缩方法 |
CN201810142763.8 | 2018-02-11 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2019153342A1 true WO2019153342A1 (zh) | 2019-08-15 |
Family
ID=62929470
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2018/076495 WO2019153342A1 (zh) | 2018-02-11 | 2018-02-12 | 一种基于增强图变换的点云属性压缩方法 |
Country Status (3)
Country | Link |
---|---|
US (1) | US11126887B2 (zh) |
CN (1) | CN108335335B (zh) |
WO (1) | WO2019153342A1 (zh) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114782564A (zh) * | 2022-06-21 | 2022-07-22 | 北京大学深圳研究生院 | 一种点云的压缩方法、装置、电子设备及存储介质 |
Families Citing this family (26)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109166160B (zh) * | 2018-09-17 | 2022-07-01 | 华侨大学 | 一种采用图形预测的三维点云压缩方法 |
CN117768658A (zh) | 2018-11-06 | 2024-03-26 | 北京字节跳动网络技术有限公司 | 依赖位置的对运动信息的存储 |
CN111327897B (zh) | 2018-12-13 | 2022-01-14 | 华为技术有限公司 | 点云编码方法和编码器 |
CN111247798B (zh) * | 2019-01-10 | 2022-03-15 | 深圳市大疆创新科技有限公司 | 对三维数据点集进行编码或解码的方法和设备 |
CN109979008B (zh) * | 2019-04-09 | 2023-04-07 | 北京大学深圳研究生院 | 一种基于属性的点云条带划分方法 |
CN111602176A (zh) * | 2019-06-03 | 2020-08-28 | 深圳市大疆创新科技有限公司 | 点云数据的位置坐标的编解码方法、系统和存储介质 |
CN111699684B (zh) * | 2019-06-14 | 2022-05-06 | 深圳市大疆创新科技有限公司 | 三维数据点的编解码方法和装置 |
KR20220029595A (ko) * | 2019-07-03 | 2022-03-08 | 광동 오포 모바일 텔레커뮤니케이션즈 코포레이션 리미티드 | 포인트 클라우드 인코딩과 디코딩 방법, 인코더, 디코더 및 컴퓨터 저장 매체 |
CN114127654B (zh) * | 2019-07-31 | 2024-01-12 | 沃尔沃卡车集团 | 形成车辆的行驶路径的方法 |
CN110572655B (zh) * | 2019-09-30 | 2023-01-10 | 北京大学深圳研究生院 | 一种基于邻居权重的参数选取和传递的点云属性编码和解码的方法及设备 |
CN110533645A (zh) * | 2019-08-23 | 2019-12-03 | 上海交通大学 | 基于视觉任务的点云数据压缩质量评价方法及系统 |
CN110569926B (zh) * | 2019-09-19 | 2021-06-08 | 北京工商大学 | 一种基于局部边缘特征增强的点云分类方法 |
WO2021062743A1 (zh) * | 2019-09-30 | 2021-04-08 | Oppo广东移动通信有限公司 | 占位信息的预测方法、编码器、解码器、及存储介质 |
EP4068789A4 (en) | 2020-01-07 | 2023-01-04 | LG Electronics Inc. | POINT CLOUD DATA TRANSMITTER DEVICE, POINT CLOUD DATA TRANSMITTER METHOD, POINT CLOUD DATA RECEIVE DEVICE AND POINT CLOUD DATA RECEIVE METHOD |
US11741637B2 (en) * | 2020-02-10 | 2023-08-29 | Tencent America LLC | Node-based geometry and attribute coding for a point cloud |
WO2021206291A1 (ko) * | 2020-04-08 | 2021-10-14 | 엘지전자 주식회사 | 포인트 클라우드 데이터 전송 장치, 전송 방법, 처리 장치 및 처리 방법 |
KR102406845B1 (ko) * | 2020-04-13 | 2022-06-10 | 엘지전자 주식회사 | 포인트 클라우드 데이터 송신 장치, 포인트 클라우드 데이터 송신 방법, 포인트 클라우드 데이터 수신 장치 및 포인트 클라우드 데이터 수신 방법 |
WO2021215811A1 (ko) * | 2020-04-24 | 2021-10-28 | 엘지전자 주식회사 | 포인트 클라우드 데이터 송신 장치, 포인트 클라우드 데이터 송신 방법, 포인트 클라우드 데이터 수신 장치 및 포인트 클라우드 데이터 수신 방법 |
CN111866518B (zh) * | 2020-07-29 | 2022-05-27 | 西安邮电大学 | 基于特征提取的自适应三维点云压缩方法 |
CN112218079B (zh) * | 2020-08-24 | 2022-10-25 | 北京大学深圳研究生院 | 一种基于空间顺序的点云分层方法、点云预测方法及设备 |
CN115474035A (zh) * | 2021-06-11 | 2022-12-13 | 鹏城实验室 | 点云属性编码方法、装置、解码方法、装置及相关设备 |
CN115474041B (zh) * | 2021-06-11 | 2023-05-26 | 腾讯科技(深圳)有限公司 | 点云属性的预测方法、装置及相关设备 |
CN117378204A (zh) * | 2021-08-12 | 2024-01-09 | Oppo广东移动通信有限公司 | 点云编解码方法、编码器、解码器及计算机可读存储介质 |
CN114339230B (zh) * | 2022-03-03 | 2022-09-02 | 杭州未名信科科技有限公司 | 用于视频编码的变换核选择方法、装置、存储介质及终端 |
CN115994849B (zh) * | 2022-10-24 | 2024-01-09 | 南京航空航天大学 | 一种基于点云上采样的三维数字水印嵌入与提取方法 |
CN115421161B (zh) * | 2022-11-03 | 2023-04-07 | 上海伯镭智能科技有限公司 | 基于激光雷达测距的无人驾驶矿车控制方法 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6529634B1 (en) * | 1999-11-08 | 2003-03-04 | Qualcomm, Inc. | Contrast sensitive variance based adaptive block size DCT image compression |
CN101005623A (zh) * | 2006-01-21 | 2007-07-25 | 宇龙计算机通信科技(深圳)有限公司 | 确定视频块帧内或帧间编码的方法 |
CN101383972A (zh) * | 2008-09-28 | 2009-03-11 | 北京航空航天大学 | 基于空间预测与变换的遥感图像压缩方法 |
CN105654530A (zh) * | 2016-03-08 | 2016-06-08 | 厦门大学 | 一种基于压缩感知的高鲁棒性图像自适应压缩方法 |
CN107403456A (zh) * | 2017-07-28 | 2017-11-28 | 北京大学深圳研究生院 | 一种基于kd树和优化图变换的点云属性压缩方法 |
Family Cites Families (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1899732A4 (en) * | 2005-05-11 | 2009-02-18 | Georgia Tech Res Inst | ACCORDABLE PLASMONIC NANOPARTICLES |
KR101791242B1 (ko) * | 2010-04-16 | 2017-10-30 | 에스케이텔레콤 주식회사 | 영상 부호화/복호화 장치 및 방법 |
CN106796661B (zh) * | 2014-08-12 | 2020-08-25 | 曼蒂斯影像有限公司 | 投影光图案的系统、方法和计算机程序产品 |
RU2560684C1 (ru) * | 2014-10-13 | 2015-08-20 | Федеральное государственное бюджетное научное учреждение "Саратовский научно-исследовательский ветеринарный институт" (ФГБНУ "Саратовский НИВИ") | Средство для диагностики лейкоза крупного рогатого скота и способ его применения |
KR102365685B1 (ko) * | 2015-01-05 | 2022-02-21 | 삼성전자주식회사 | 인코더의 작동 방법과 상기 인코더를 포함하는 장치들 |
US10013653B2 (en) * | 2016-01-26 | 2018-07-03 | Università della Svizzera italiana | System and a method for learning features on geometric domains |
EP3249921A1 (en) * | 2016-05-24 | 2017-11-29 | Thomson Licensing | Method, apparatus and stream for immersive video format |
EP3429207A1 (en) * | 2017-07-13 | 2019-01-16 | Thomson Licensing | A method and apparatus for encoding/decoding a colored point cloud representing the geometry and colors of a 3d object |
US11113845B2 (en) * | 2017-09-18 | 2021-09-07 | Apple Inc. | Point cloud compression using non-cubic projections and masks |
US10607373B2 (en) * | 2017-11-22 | 2020-03-31 | Apple Inc. | Point cloud compression with closed-loop color conversion |
-
2018
- 2018-02-11 CN CN201810142763.8A patent/CN108335335B/zh active Active
- 2018-02-12 US US16/957,187 patent/US11126887B2/en active Active
- 2018-02-12 WO PCT/CN2018/076495 patent/WO2019153342A1/zh active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6529634B1 (en) * | 1999-11-08 | 2003-03-04 | Qualcomm, Inc. | Contrast sensitive variance based adaptive block size DCT image compression |
CN101005623A (zh) * | 2006-01-21 | 2007-07-25 | 宇龙计算机通信科技(深圳)有限公司 | 确定视频块帧内或帧间编码的方法 |
CN101383972A (zh) * | 2008-09-28 | 2009-03-11 | 北京航空航天大学 | 基于空间预测与变换的遥感图像压缩方法 |
CN105654530A (zh) * | 2016-03-08 | 2016-06-08 | 厦门大学 | 一种基于压缩感知的高鲁棒性图像自适应压缩方法 |
CN107403456A (zh) * | 2017-07-28 | 2017-11-28 | 北京大学深圳研究生院 | 一种基于kd树和优化图变换的点云属性压缩方法 |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114782564A (zh) * | 2022-06-21 | 2022-07-22 | 北京大学深圳研究生院 | 一种点云的压缩方法、装置、电子设备及存储介质 |
CN114782564B (zh) * | 2022-06-21 | 2022-09-09 | 北京大学深圳研究生院 | 一种点云的压缩方法、装置、电子设备及存储介质 |
Also Published As
Publication number | Publication date |
---|---|
CN108335335A (zh) | 2018-07-27 |
CN108335335B (zh) | 2019-06-21 |
US11126887B2 (en) | 2021-09-21 |
US20200394450A1 (en) | 2020-12-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2019153342A1 (zh) | 一种基于增强图变换的点云属性压缩方法 | |
WO2019153326A1 (zh) | 一种基于帧内预测的点云属性压缩方法 | |
WO2019213986A1 (zh) | 一种基于多角度自适应帧内预测的点云属性压缩方法 | |
US10552989B2 (en) | Point cloud attribute compression method based on KD tree and optimized graph transformation | |
Huang et al. | Octree-Based Progressive Geometry Coding of Point Clouds. | |
WO2021022621A1 (zh) | 一种基于邻居的权重优化的点云帧内预测方法及设备 | |
WO2019213985A1 (zh) | 一种基于层次划分的点云属性压缩方法 | |
Golla et al. | Real-time point cloud compression | |
CN108875813B (zh) | 一种基于几何图像的三维网格模型检索方法 | |
JP7330306B2 (ja) | 変換方法、逆変換方法、エンコーダ、デコーダ及び記憶媒体 | |
JP7386337B2 (ja) | 分割方法、符号器、復号器及びコンピュータ記憶媒体 | |
CN105141970B (zh) | 一种基于三维模型几何信息的纹理图像压缩方法 | |
Pavez et al. | Dynamic polygon cloud compression | |
CN113518226A (zh) | 一种基于地面分割的g-pcc点云编码改进方法 | |
CN111598786A (zh) | 一种基于深度去噪自编码网络的高光谱图像解混方法 | |
US20230237705A1 (en) | Methods for level partition of point cloud, and decoder | |
CN104299256B (zh) | 一种近似无损的三维体数据压缩域体绘制方法 | |
CN112488117A (zh) | 一种基于方向诱导卷积的点云分析方法 | |
Chen et al. | Point cloud attribute compression via successive subspace graph transform | |
CN113763539B (zh) | 一种基于图像和三维输入的隐式函数三维重建方法 | |
CN105205496A (zh) | 基于空间信息约束的增强型稀疏表示高光谱图像分类装置及方法 | |
WO2022067782A1 (zh) | 一种点云数据的层次划分方法、编码器及存储介质 | |
CN118212278B (zh) | 一种基于卷积的图像配准方法 | |
WO2023245981A1 (zh) | 一种点云的压缩方法、装置、电子设备及存储介质 | |
CN117321996A (zh) | 帧内预测、编解码方法及装置、编解码器、设备、介质 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 18905040 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 18905040 Country of ref document: EP Kind code of ref document: A1 |