Attribute-guided point cloud attribute compression method based on graph transformation

By introducing attribute information to assist in Laplace matrix establishment and graph transformation, the point cloud attribute compression method is optimized, which solves the problem that geometric information cannot accurately reflect attribute correlation in point cloud attribute compression, and achieves more efficient point cloud attribute compression.

WO2025179573A1PCT designated stage Publication Date: 2025-09-04PEKING UNIV SHENZHEN GRADUATE SCHOOL
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2024/079524
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2024-03-01
Publication Date
2025-09-04

AI Technical Summary

Technical Problem

In the prior art, the graph-based point cloud attribute compression method obtains an adjacency matrix using geometric information cannot accurately reflect the attribute correlation between points, resulting in limited compression performance.

Method used

By introducing attribute information to assist in the establishment of the Laplace matrix, K-D tree block division and graph transformation are used, and attribute clustering and entropy coding are combined to optimize the graph transformation process and improve the compression efficiency of point cloud attributes.

Benefits of technology

Improve the performance of point cloud attribute compression, reduce data redundancy, and improve encoding and decoding efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2024079524_04092025_PF_FP_ABST
    Figure CN2024079524_04092025_PF_FP_ABST
Patent Text Reader

Abstract

The present invention provides an attribute-guided point cloud attribute compression method based on graph transformation, comprising the following steps: S1, preprocessing: preprocessing original point cloud data; S2, K-D tree block division; S3, block-wise graph transformation; S4, attribute classification flag bit encoding; S5, transform coefficient processing; S6, entropy encoding: using an arithmetic encoder to perform lossy encoding on a quantized alternating current coefficient; and S7, point cloud reconstruction: a decoding end performing entropy decoding on a bitstream obtained by the entropy encoding, performing inverse quantization to obtain a transform coefficient, performing reconstruction on the basis of known geometric information of the decoding end, and realizing inverse transformation on the basis of attribute classification obtained by decoding, to reconstruct attribute information of a point cloud. The method of the present invention uses graph structure establishment based on geometric information as a basis and appropriately introduces attribute information, thereby efficiently compressing point cloud data, and reducing redundant information of data.
Need to check novelty before this filing date? Find Prior Art

Description

An attribute-guided point cloud attribute compression method based on graph transformation Technical Field

[0001] The present invention relates to the field of point cloud compression, and in particular to an attribute-guided point cloud attribute compression method based on graph transformation. Background Art

[0002] Due to growing commercial demand, three-dimensional point clouds are becoming increasingly popular in academia and industry. Three-dimensional point clouds are often obtained through radar scanning and typically contain two types of information: geometry and attribute information. Attribute information can include color, reflectivity, normal vectors, and more. This rich information makes point clouds extremely large in size, so, like image compression, point clouds also require compression. Unlike image compression, the irregularity and disorder of point clouds mean that point cloud compression involves both geometry and attribute tasks. In most point cloud compression methods, geometry and attribute compression are independent of each other, and attribute encoding and decoding are often performed when the geometric information is known. The graph Fourier transform, due to its flexible adjacency matrix, is highly adaptable to irregular signals and therefore excels in point cloud attribute compression.

[0003] The main problem in the existing technology is that the mainstream graph-based point cloud attribute compression method uses geometric information to derive the adjacency matrix, but its performance is limited because the geometric relationship cannot well reflect the correlation of attributes in some cases.

[0004] The difficulty in solving the above problems and improving the performance of point cloud encoding and decoding lies in how to make the Laplacian matrix in the graph transformation more accurately express the correlation between the attributes of the points.

[0005] The significance of solving this problem lies in adopting a clever method to appropriately introduce attribute information, assist in the establishment of the Laplacian matrix, improve the efficiency of graph transformation and thus enhance the performance of attribute compression.

[0006] Disclosure of the invention

[0007] The present invention provides an attribute-guided point cloud attribute compression method based on graph transformation, which is based on the establishment of a graph structure based on geometric information and appropriately introduces attribute information. It can efficiently compress point cloud data and reduce redundant information in the data.

[0008] The technical solutions of the present invention are as follows:

[0009] The attribute-guided point cloud attribute compression method based on graph transformation of the present invention comprises the following steps: S1. Preprocessing: preprocessing the original point cloud data, including geometric preprocessing, geometric encoding and decoding, and attribute color space conversion; S2. KD tree block partitioning: dividing the preprocessed point cloud into a plurality of small block structures with similar bounding box sizes and close to cubes by the KD tree partitioning method; S3. Graph transformation based on blocks: using the geometric distance between points as the standard for establishing weights, constructing a graph structure composed of all points in the block, guiding the establishment of the graph Laplacian matrix by clustering the attributes of the points in the block, and using the eigenvector matrix of the Laplacian matrix as the transformation matrix to transform the points in the block. Point attributes are transformed; S4. Attribute classification flag encoding: For each block, one bit is used to record the attribute classification in a circuitous scanning manner, and a two-dimensional wavelet transform is performed to reduce the bit rate; S5. Transform coefficient processing: The transform coefficient is divided into two parts, DC and AC, and different methods are used to process them before entropy coding; S6. Entropy coding: For the quantized AC coefficients, an arithmetic encoder is used to perform lossy coding; and S7. Point cloud reconstruction: The decoding end entropy decodes the bit stream obtained by entropy coding, dequantizes to obtain transform coefficients, reconstructs the map based on the geometric information known to the decoding end, and implements inverse transformation with the help of the attribute classification obtained by decoding to reconstruct the attribute information of the point cloud.

[0010] Optionally, in the above-mentioned attribute-guided point cloud attribute compression method based on graph transformation, in step S3, the geometric distances between all points in a point cloud are first calculated, a weight matrix is ​​determined according to the rule that the higher the geometric distance, the smaller the correlation, and the weight matrix is ​​used to calculate the Laplace matrix; the eigenvectors of the Laplace matrix are obtained and a new matrix is ​​formed, which is used as the transformation matrix to perform linear transformation on the attributes of the point cloud piece.

[0011] Optionally, in the above-mentioned attribute-guided point cloud attribute compression method based on graph transformation, after establishing the weight matrix, the attributes of the points in the block are first divided into two categories using spectral clustering, and the weights between any two points belonging to different categories are reset to zero, so that the block is divided into two sub-blocks, and graph transformation is performed separately.

[0012] Optionally, in the above-mentioned attribute-guided point cloud attribute compression method based on graph transformation, in step S5, the transformation coefficients obtained after the transformation are divided into two categories: DC coefficients and AC coefficients. The DC coefficient is the first coefficient of each block. The DC coefficient is extracted, and the geometric center of gravity of each block is used as its corresponding geometric coordinate to form a new point cloud. The inter-block correlation is further utilized and the existing predictive transformation technology is used to directly perform compression; the AC coefficients of different blocks are divided into four categories according to the number of non-zero values ​​of the quantized transformation coefficients, and are truncated with different lengths and sent to the entropy encoder.

[0013] Optionally, in the above-mentioned attribute-guided graph transformation-based point cloud attribute compression method, in step S6, the quantized AC coefficients of each block are spliced ​​together to form a matrix with a dimension of the total number of point cloud points × the number of attribute channels, and are losslessly encoded using an arithmetic encoder.

[0014] According to the technical solution of the present invention, the beneficial effects produced are:

[0015] Based on the correlation between point cloud geometry and attribute information and the correlation between attribute information between points, the present invention proposes an attribute-guided point cloud attribute compression method based on graph transformation. The method of the present invention adopts the traditional signal processing idea, mainly utilizing the high efficiency of graph transformation for irregular signal compression to compress the attribute values ​​of the point cloud (taking color as an example). The traditional graph transformation method approximates the point cloud as a Markov random field, and represents the attribute difference representing the attribute correlation with geometric information. However, since geometry and attributes are not strongly correlated, this method has an upper limit. The method of the present invention improves and optimizes this problem by introducing attribute information by dividing the block into sub-blocks, so that the introduced attribute information does not bring too high additional coding overhead, and at the same time enables the Laplacian matrix to more accurately express the correlation between points in the block, thereby improving the efficiency of the graph transformation; at the same time, the present invention classifies the transformation coefficients and optimizes the attribute coding based on graph transformation as a whole.

[0016] In order to better understand and illustrate the concept, working principle and effect of the present invention, the present invention is described in detail below with reference to the accompanying drawings and through specific embodiments:

[0017] BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific implementation of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the specific implementation or the description of the prior art.

[0019] FIG1 is a flow chart of an attribute-guided point cloud attribute compression method based on graph transformation according to the present invention;

[0020] FIG2 is an overview of the coding framework structure of the present invention;

[0021] FIG3 is a schematic diagram of block partitioning based on the KD tree;

[0022] FIG4 is a flowchart of graph transformation with attribute introduction;

[0023] FIG5 is a schematic diagram of transform coefficient processing;

[0024] Figure 6 shows the geometric classification of points within a block after clustering by attributes;

[0025] FIG7 is a visual comparison of the attribute classification flag scanning order of the present invention and the traditional Morton order;

[0026] FIG8 shows the traversal order of the three dimensions of the scan attribute flag.

[0027] Best Mode for Carrying Out the Invention

[0028] To make the purpose, technical methods and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings and specific examples. These examples are merely illustrative and not intended to limit the present invention.

[0029] The demand for efficient storage and transmission of point clouds has driven the development of point cloud compression algorithms. The present invention's attribute-guided graph-transformation-based point cloud attribute compression method involves a point cloud encoding and decoding method using transforms and frequency domain analysis. It aims to encode the attributes of the input point cloud into a compact bitstream that can be successfully decoded and reconstructed at the decoding end to reconstruct the attributes of the original point cloud.

[0030] As shown in FIG1 , the attribute-guided point cloud attribute compression method based on graph transformation of the present invention includes the following steps:

[0031] S1. Preprocessing: Preprocess the original point cloud data, including geometric preprocessing, geometric encoding and decoding, attribute color space conversion and other operations to facilitate subsequent encoding steps.

[0032] S2. KD Tree Block Partitioning: The preprocessed point cloud is partitioned into multiple small blocks with similarly sized, cube-like bounding boxes using a KD tree. Specifically, the bounding box of the preprocessed point cloud is partitioned into multiple small blocks with similar sizes and cube-like bounding boxes using a KD tree structure. Each block contains a segmented point cloud, as shown in Figure 3.

[0033] S3. Block-based graph transformation: Using the geometric distance between points as the criterion for establishing weights, a graph structure consisting of all points in a block is constructed. The attributes of the points in the block are clustered to guide the establishment of the graph Laplacian matrix, and the eigenvector matrix of the Laplacian matrix is ​​used as the transformation matrix to transform the attributes of the points in the block.

[0034] Specifically, the graph transformation is performed on the point cloud slices in each block, as shown in Figure 4. First, the geometric distance between all points in a point cloud is calculated, and the weight matrix is ​​determined according to the rule that the higher the geometric distance, the smaller the correlation, and the weight matrix is ​​used to calculate the Laplace matrix. The eigenvectors of the Laplace matrix are obtained and a new matrix is ​​formed, which is used as the transformation matrix to perform a linear transformation on the attributes of the point cloud slice. In order to be able to effectively extract the DC coefficient in subsequent operations, it is necessary to ensure that the established graph structure does not contain multiple independent subgraphs. Among them, after the weight matrix is ​​established, the attributes of the points in the block are first divided into two categories using spectral clustering (as shown in Figure 6). The clustering groups are encoded by a 1-bit flag bit, and then the weights between any two points belonging to different categories are reset to zero, so that the block is divided into two sub-blocks, and the graph transformation is performed separately. The transformation coefficients are quantized to obtain quantized coefficients.

[0035] S4. Attribute classification flag encoding: For each block, use one bit to record the attribute classification in a circuitous scanning manner, and perform a two-dimensional wavelet transform on it to reduce the bit rate.

[0036] Specifically, the classification of each block's attributes is recorded in a separate bitstream. Since the attributes are divided into two categories, a single bit is sufficient to record the classification flag. Points belonging to different categories are distinguished by 0s and 1s, and the flags are recorded as a stream using a zigzag scan (as shown in Figure 7 (b)). A two-dimensional wavelet transform is used to extract transitions, further improving compression efficiency.

[0037] S5. Transform coefficient processing: The transform coefficients are divided into two parts, DC and AC, and processed in different ways before entropy coding.

[0038] Specifically, the transformed coefficients are divided into two categories: DC coefficients and AC coefficients, as shown in Figure 5. The DC coefficient is the first coefficient of each block. The DC coefficient is extracted and its corresponding geometric coordinates are taken from the geometric center of gravity of each block to form a new point cloud. This is then directly compressed using existing predictive transform techniques, leveraging inter-block correlations. The AC coefficients of different blocks are divided into four categories (cat1, cat2, cat3, and cat4 in Figure 5) based on the number of non-zero values ​​in the quantized transform coefficients. These coefficients are then truncated to different lengths and fed into the entropy encoder.

[0039] S6. Entropy Coding: Lossy encoding of the quantized AC coefficients using an arithmetic encoder to achieve efficient compression.

[0040] Specifically, the quantized AC coefficients of each block are concatenated together to form a matrix with the dimension of the total number of point cloud points × the number of attribute channels, which is lossily encoded using an arithmetic encoder to achieve efficient compression.

[0041] S7. Point cloud reconstruction: The decoder performs entropy decoding on the bitstream obtained by entropy coding, dequantizes the transformed coefficients, and reconstructs the map based on the geometric information known to the decoder. The decoded attribute classification is used to perform the inverse transformation and reconstruct the attribute information of the point cloud.

[0042] The following describes the relevant steps in detail with reference to the specific implementation steps in FIG. 2 to FIG. 8 .

[0043] S1. Preprocessing

[0044] Preprocessing includes geometric preprocessing, including quantization, coordinate transformation, and voxelization, which facilitates subsequent encoding steps. Geometric encoding and decoding means that in the entire attribute encoding and decoding process, geometry is always known information and can be called at any time. Attribute color space conversion refers to converting the color attributes of the point cloud from RGB space to YUV space, removing the correlation between color channels to a certain extent. Color space conversion is achieved by linear transformation, and the formula is:

[0045] Among them, a is the color value of a point in RGB space, and a′ is the color value converted to YUV space.

[0046] S2.KD tree block partitioning

[0047] The block partitioning method based on the KD tree can make the shape and size of the final divided blocks relatively uniform by customizing the dimension of each partition. Compared with the conventional KD tree construction algorithm, the present invention does not store points at non-leaf nodes, but divides the point cloud into two blocks evenly according to the principle of equality (the number of points in the left child node is greater when the number is odd). At the same time, the node division is not carried out to the end, but stops when a certain number of layers is reached. The number of division layers is determined by the approximate number of points expected to be contained in each block. Regarding the principle of selecting the dimension for each node division, in order to ensure that the final divided blocks are evenly distributed in the three dimensions of X, Y, and Z, therefore, at each partition, the variance of the points contained in the node in the three dimensions of X, Y, and Z is calculated to compare the degree of discreteness in the three dimensions, and finally the dimension with the highest degree of discreteness is selected for division: d cut =max(v x ,v y ,v z )

[0048] Among them, d cut is the dimension used for each partition, v x ,v y ,v z Represents the variance of all points in a block in three geometric dimensions.

[0049] The effects before and after block division are shown in Figure 3.

[0050] S3. Block-based graph transformation

[0051] In graph transformation, the geometric information of the point cloud is used to construct an adjacency matrix, which exploits the spatial correlation of signals observed in nature. The attribute correlation between two points is approximated by their geometric distance. The weight matrix W is defined as follows:

[0052] Among them, D i,j defines the distance between vertices i and j in the graph, and δ defines D i,j The absolute median difference of τ defines the minimum distance between vertices based on the assumption that the graph signal is regarded as a Gaussian Markov Random Field (GMRF).

[0053] In order to facilitate the subsequent extraction of DC coefficients, it is necessary to ensure that the graph structure established for each block does not include independent subgraphs. Before attribute clustering, there are usually multiple independent subgraphs after the weight matrix is ​​established based on geometric information. This paper assumes that for any two subgraphs, G1 and G2, let v i and v j are the two points with the shortest geometric distance in the subgraph. The weight matrix will be updated as follows:

[0054] This means that W i,j Not affected by the threshold. Once two subgraphs are connected, they can be treated as a single graph, which can then be further merged with the next subgraph, and the process continues until only one graph remains within the entire block / sub-block.

[0055] Based on the above graph transformation method, the present invention divides the points within a block into two clusters by performing spectral clustering on the attributes. This is shown in Figure 6. This secondary partitioning of the block based on the attributes is equivalent to removing the edge between two points with significantly different attributes. Adjustments to the weight matrix can be expressed as resetting the weights of points between any two clusters to zero.

[0056] S4. Attribute classification flag encoding

[0057] In the present invention, the number of attribute classes, or sub-blocks, is set to 2, allowing each point to use an additional symbol (1 bit) to represent the attribute class flag. A special scanning method is designed for attribute classification. Considering that points belonging to the same class within a block are often spatially clustered, the present invention uses an appropriate scanning method to ensure that flags with the same value appear concentrated in the resulting sequence. This significantly improves the efficiency of the arithmetic encoder by reducing the variation in the classification flags. To minimize symbol discontinuities during the scanning process, the present invention selects the direction with the most consecutive points as the primary scanning direction, that is, the direction with the fewest points in the top view. The scanning order is determined based on the number of points in the top view (as shown in Figure 8, the first layer of the top view plane is traversed, followed by the second and third layers). Subsequently, a zigzag scanning method (zigzag pattern) is used to scan all points in the entire block, as shown in Figure 7. Figure 7 (a) shows the flags scanned according to the original point cloud sorting method, namely the Morton order, while (b) is the more concise method adopted by the present invention. After scanning the code stream of the flag bit, two-dimensional wavelet transform is applied to it to extract the changes of the flag bit in the code stream. The scanning order of the present invention improves the coding efficiency because it introduces fewer changes.

[0058] S5. Transformation coefficient processing

[0059] After quantization with a specific quantization step size, the transformed transform coefficients are divided into two parts: DC coefficients and AC coefficients, and processed separately. Since the previous steps ensure that each block or sub-block contains only one independent graph structure, it is guaranteed that the first bit of each block transform coefficient can represent the only DC coefficient in this block or sub-block. The calculation formula of the DC coefficient can be expressed as:

[0060] Where f is the image signal, that is, the point cloud attribute, and n is the number of points in the block. Its relationship with the average attribute value of the points in the block is:

[0061] The present invention extracts the DC coefficients and combines them with the geometric center of gravity of each block to form a smaller point cloud with lower resolution, as shown in Figure 5. This new point cloud is then compressed using existing predictive transformation technology to achieve encoding of the DC coefficients.

[0062] When encoding AC coefficients, the transform coefficients obtained through graph transformation typically have energy concentrated in low-frequency components. Therefore, after quantization, the latter portion of the coefficient sequence often contains a large number of zero values, while the non-zero values ​​are primarily ones. The present invention omits encoding this portion, achieving significant bitrate gains with only a slight loss in PSNR. Therefore, after quantization, the present invention classifies the AC coefficients of each block into four categories based on the distribution of non-zero coefficients. A specific length is then truncated from the end of the sequence based on the category, as shown in Figure 5. This approach achieves bitrate savings by reducing coefficient length.

[0063] S6. Entropy Coding

[0064] The quantized AC coefficients of each block are concatenated to form a matrix with dimensions equal to the total number of point cloud points × the number of attribute channels. This matrix is ​​then lossily encoded using an arithmetic encoder for efficient compression. The final bitrate can be adjusted by controlling the quantization step size. A higher quantization step size results in a lower bitrate but also higher distortion.

[0065] S7. Point cloud reconstruction

[0066] The decoder performs entropy decoding on the bitstream obtained by entropy coding to restore it to the state of a matrix of the total number of point cloud points × the number of attribute channels. It then dequantizes it to obtain the transformation coefficients. The geometric information is known at the decoder, and the same method as the encoder is used for mapping. The attributes obtained by decoding are classified to obtain the same Laplacian matrix as the encoder, and the inverse transformation is implemented to reconstruct the attribute information of the point cloud. Finally, the color attributes are converted from the YUV color space back to the RGB color space.

[0067] Table 1 Comparison of compression performance between the present invention and other methods

[0068] The framework proposed in this paper comprehensively optimizes the graph-transformation-based point cloud attribute compression framework. It introduces an attribute-assisted graph transformation process as its core, and adds a module for separate processing of DC and AC coefficients. This maximizes the correlation between blocks after segmentation and the correlation between points within a block, thereby improving compression efficiency. Experimental data from this paper is shown in Table 1. The point cloud compression performance metric is BD-rate. BD-rate (also known as BDBR) is defined as the average difference between two RD curves, which is in turn defined as the difference between the area integral of the lower curve divided by the integral interval and the area integral of the upper curve divided by the integral interval. BD-rate represents the rate increase of the optimized algorithm compared to the original algorithm, while maintaining the same objective quality. A negative BD-rate indicates improved coding performance of the optimized algorithm. The experimental dataset used four MPEG8i test sequences. Table 1 compares the proposed method with various methods, including MPEG's two latest G-PCC codecs (RAHT and PLT) and the basic graph-based attribute compression framework HAC. The overall compression performance of the proposed method is superior to these methods.

[0069] The above description is the best embodiment based on the concept and working principle of the invention. The above embodiment should not be understood as limiting the scope of protection of the present claims. Other embodiments and combinations of implementations of the present invention are also within the scope of protection of the present invention. Industrial Applicability

[0070] The present invention provides an attribute-guided point cloud attribute compression method based on graph transformation, which can be applied in various fields, especially in computer vision, condition monitoring, virtual reality, robotics and autonomous driving.

Claims

1. An attribute-guided point cloud attribute compression method based on graph transformation, characterized in that: The following steps are involved: S1. Preprocessing: Preprocessing the original point cloud data, including geometric preprocessing, geometric encoding and decoding, and attribute color space conversion; S2. KD tree block partitioning: The preprocessed point cloud is partitioned into multiple small block structures with similar bounding box sizes and close to cubes using the KD tree partitioning method; S3. Block-based graph transformation: Using the geometric distance between points as the criterion for weighting, a graph structure consisting of all points within a block is constructed. Clustering the attributes of the points within the block guides the construction of a graph Laplacian matrix. The eigenvector matrix of the Laplacian matrix is ​​used as the transformation matrix to transform the attributes of the points within the block. S4. Attribute classification flag encoding: For each block, a circuitous scanning method is used to record the attribute classification, and a two-dimensional wavelet transform is performed to reduce the bit rate; S5. Transform coefficient processing: The transform coefficients are divided into two parts, DC and AC, and processed in different ways before entropy coding; S6. Entropy coding: For the quantized AC coefficients, use an arithmetic encoder to perform lossy encoding; as well as S7. Point cloud reconstruction: The decoder performs entropy decoding on the bitstream obtained by entropy coding, dequantizes the transformed coefficients, and reconstructs the map based on the geometric information known to the decoder. The decoded attribute classification is used to perform the inverse transformation and reconstruct the attribute information of the point cloud.

2. The attribute-guided point cloud attribute compression method based on graph transformation according to claim 1, characterized in that: In step S3, the geometric distances between all points in a point cloud are first calculated, and a weight matrix is ​​determined according to the rule that the higher the geometric distance, the smaller the correlation. The weight matrix is ​​used to calculate the Laplace matrix. The eigenvectors of the Laplace matrix are obtained and formed into a new matrix, which is used as the transformation matrix to perform linear transformation on the attributes of the point cloud.

3. The attribute-guided point cloud attribute compression method based on graph transformation according to claim 2, characterized in that: in, After establishing the weight matrix, the attributes of the points in the block are first divided into two categories using spectral clustering, and the weights between any two points belonging to different categories are reset to zero, so that the block is divided into two sub-blocks, and graph transformation is performed separately.

4. The attribute-guided point cloud attribute compression method based on graph transformation according to claim 1, characterized in that: In step S5, the transformation coefficients obtained after the transformation are divided into two categories: DC coefficients and AC coefficients. The DC coefficient is the first coefficient of each block. The DC coefficient is extracted, and the geometric center of gravity of each block is used as its corresponding geometric coordinate to form a new point cloud. The inter-block correlation is further utilized and the existing predictive transformation technology is used to directly compress it; the AC coefficients are divided into four categories according to the number of non-zero values ​​of the quantized transformation coefficients, and are truncated with different lengths and sent to the entropy encoder.

5. The attribute-guided point cloud attribute compression method based on graph transformation according to claim 1, characterized in that: In step S6, the quantized AC coefficients of each block are concatenated together to form a matrix with a dimension of the total number of point cloud points × the number of attribute channels, and lossy encoding is performed using an arithmetic encoder.

Citation Information

Patent Citations

  • KD tree and graph transformation optimization-based point cloud attribute compression method

    CN107403456A

  • Enhanced diagram conversion-based point cloud attribute compression method

    CN108335335A

  • Point cloud attribute compression method for graph sparsity optimization based on texture driving

    CN117319683A

  • Point cloud data transmission device, point cloud data transmission method, point cloud data reception device and point cloud data reception method

    US20220286713A1