A point cloud quality evaluation method based on relative entropy

Abstract

The present application relates to a kind of point cloud quality evaluation methods based on relative entropy, belong to multimedia information processing field, reference point cloud is down-sampled, and the geometric architecture of point cloud is obtained;Sub-point cloud is constructed in reference point cloud and distorted point cloud with geometric architecture as center, its quick point feature histogram and its relative entropy are calculated, and are taken as the geometric quality result of sub-point cloud;The color attribute of sub-point cloud is converted to LMN color space, the zeroth moment of color space, the first moment, the second moment are calculated, and the color quality result of sub-point cloud is obtained;The geometric quality result and color quality result of sub-point cloud are aggregated with weight, and the final point cloud quality evaluation model based on relative entropy is obtained.

Description

Technical Field

[0001] This invention belongs to the field of multimedia information processing and relates to a point cloud quality assessment method based on relative entropy. Background Technology

[0002] Point clouds, as an emerging multimedia type, have received widespread attention in fields such as intelligent transportation, autonomous driving, and virtual reality. A typical point cloud consists of millions of unstructured, non-uniformly distributed 3D points, each of which contains at least positional and color information.

[0003] Similar to traditional image processing methods, point cloud processing involves steps such as sampling, compression, transmission, reconstruction, rendering, and analysis. Due to technological limitations, the quality of point clouds will inevitably be compromised during these processes. For example, lossy compression can lead to distortion, and sampling can be affected by noise contamination. Therefore, accurate quality assessment of lossy point clouds is crucial.

[0004] Point cloud quality assessment can be divided into subjective assessment and objective assessment. Subjective assessment evaluates point clouds based on the viewer's subjective perception. A common subjective quality assessment method provides an original point cloud (reference point cloud) and a distorted point cloud (the point cloud to be assessed), allowing viewers to rate the image to be assessed. The average of all subjective scores is then calculated. Objective assessment uses a specific mathematical model to quantify the difference between the reference image and the image to be assessed. Subjective assessment is influenced by factors such as viewer preferences, viewing equipment, and location, resulting in high costs and low efficiency. Objective assessment scores are unaffected by subjective factors and are widely used in the field of point cloud quality assessment. The applicant found that existing objective point cloud quality assessment algorithms lack sufficient accuracy, leading to inefficiencies in subsequent processing steps. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a point cloud quality assessment method based on relative entropy. This method considers both the geometric and color features of the point cloud. First, geometric key points are extracted by resampling a reference point cloud to form a geometric skeleton. Then, sub-point clouds are constructed centered on these key points in both the reference and distorted point clouds. The sub-point clouds of the reference and distorted point clouds are evaluated from two perspectives: geometric distortion and color distortion. First, fast point feature histograms of the sub-point clouds of the reference and distorted point clouds are calculated. Then, the relative entropy between the point feature histograms is calculated to measure the degree of geometric distortion, thus obtaining the geometric quality of the sub-point clouds. The statistical moments of the color gradients of the sub-point clouds of the reference and distorted point clouds are calculated to measure the degree of color distortion, thus obtaining the color quality of the sub-point clouds. Finally, the geometric and color qualities of all sub-point clouds are linearly weighted to obtain a tensor point cloud quality assessment model based on relative entropy.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A point cloud quality assessment method based on relative entropy includes the following steps:

[0008] S1: Downsample the reference point cloud and obtain the geometric structure of the downsampled reference point cloud through a high-pass graph filter.

[0009] S2: Using the k-nearest neighbor classification algorithm, construct sub-point clouds centered on the geometric architecture in the reference point cloud and the distorted point cloud, denoted as S1 and S2 respectively. r and S d ;

[0010] S3: Calculate S r and S d Fast point feature histogram F r and F d Then calculate F r and F d The relative entropy is used as the geometric quality result of the sub-point cloud;

[0011] S4: [S] r and S d The color attributes are converted from the RGB color space to the LMN color space, the zeroth moment, the first moment, and the second moment of the color space are calculated, and the color quality result of the sub-point cloud is obtained by weighted calculation.

[0012] S5: The geometric and color quality results of the weighted aggregate sub-point clouds are used to obtain the final tensor point cloud quality assessment model based on relative entropy.

[0013] Furthermore, step S1 includes the following steps:

[0014] S11: The reference point cloud consists of N points, each point has K attributes, denoted as:

[0015]

[0016] in, This represents the points from the 1st to the Nth. Represents a space of dimension N×K;

[0017] S12: Use geometric distance To calculate the midpoint of the sub-point cloud and The connection weights between them are defined as follows:

[0018]

[0019] Where, σ 2denoted by , where τ represents the variance between points, and τ represents the Euclidean distance threshold for clustering adjacent points into the same sub-point cloud.

[0020] S13: Use a diagonal matrix D to measure the edge density of each vertex. Connection density The edge density D is defined as follows:

[0021]

[0022] S14: The high-pass filter through which the reference point cloud passes is represented as:

[0023]

[0024] Where L is the filter length, h l It is the l-th filter coefficient; It is a point cloud shifting operator, defined as:

[0025] A = D -1 W

[0026] S15: Based on h(A), select a Haar-like local domain filter h. HH (A) Implement a high-pass filter, defined as follows:

[0027]

[0028] Where λ1~λ N Let U be the eigenvalue of A, and U be the eigenvector of A.

[0029] S16: Each point that passes through the filter The frequency response is:

[0030]

[0031] in Point A set of neighboring points, Point and The characteristic covariance matrix;

[0032] Let the filter sampling frequency be f. s Total number of points in the sampled point cloud Resampled reference point cloud Each point in it is a geometric structure.

[0033] Furthermore, in step S2, given a Euclidean distance threshold γ, the distance between the reference point cloud and the distorted point cloud is... Geometric key points in A sub-point cloud is constructed using the k-nearest neighbor search algorithm centered at the point, defined as follows:

[0034]

[0035]

[0036] Where S r This represents a sub-point cloud constructed within a reference point cloud, centered on a geometric architecture. This refers to points in a reference point cloud. S represents the reference point cloud after resampling. d This represents a sub-point cloud constructed from a distorted point cloud, centered on a geometric architecture. This represents points in a distorted point cloud. This represents the distorted point cloud after resampling;

[0037] The points in the sub-point cloud are arranged in ascending order of their distance from the geometric key point.

[0038] Furthermore, the specific steps in step S3 of treating the relative entropy as a geometric quality result of the sub-point cloud include:

[0039] S31: From S r Selecting distance geometric key points Recent k r Points, this k r The maximum distance between a point and the geometric key point is denoted as .

[0040] From S d Select the k closest to the geometric key point. d Points, this k d The maximum distance between a point and the geometric key point is denoted as .

[0041] S32: In S r In the middle, calculation Each pair within the distance and its neighboring points Point feature histogram Figure 3 elements Then, the statistics are compiled into a simple point feature histogram.

[0042] In S d In the middle, calculation Each pair within the distance and its neighboring points Point feature histogram Figure 3 elements Then, the statistics are compiled into a simple point feature histogram.

[0043] S33: In S rIn the middle, determine each point exist Find the neighboring points within the distance and calculate each one. Point feature histogram Figure 3 elements Then, the statistics are compiled into a simple point feature histogram.

[0044] In S d In the middle, determine each point exist Histogram of features of points within the distance Figure 3 elements Then, the statistics are compiled into a simple point feature histogram.

[0045] S34: Calculate S r and S d Fast Point Feature Histogram (FPFH) r and FPFH d The definition is as follows:

[0046]

[0047]

[0048] Where d i It is the Euclidean distance between points;

[0049] S35: Calculate FPFH r and FPFH d The relative entropy between them yields the geometric quality assessment index D. i The relative entropy between histograms is defined as:

[0050]

[0051]

[0052] Furthermore, the calculation of the color quality result of the sub-point cloud in step S4 specifically includes the following steps:

[0053] S41: [S] r and S d Color attributes are converted from RGB space to LMN space:

[0054]

[0055] S42: Define the zeroth moment, the first moment, and the second moment as the total number of moments m. g Mean μ g σ g ;

[0056] S43: Define S r and S d The three measures of the color attribute are:

[0057]

[0058]

[0059]

[0060] N0, N1, and N3 are three small non-zero numbers to prevent the denominator from being 0;

[0061] S44: Multiply the above three metrics together to get S r and S d Quality measurement results of color attributes:

[0062]

[0063] Furthermore, the geometric and color quality results of the weighted aggregated sub-point clouds described in step S5 are used to obtain the final tensor point cloud quality assessment model based on relative entropy:

[0064]

[0065] in, and These are the weighting factors for the geometric quality assessment index and the color quality assessment index, respectively.

[0066] The beneficial effects of this invention are that this model not only evaluates the color quality of point clouds, but also accurately evaluates the geometric tensor of point clouds based on the theory of relative entropy.

[0067] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0068] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0069] Figure 1 This is a system flowchart used in this invention;

[0070] Figure 2 This is a schematic diagram of point cloud resampling in step S1 of the method of the present invention;

[0071] Figure 3This is a schematic diagram of the clustering based on the geometric architecture in step S2 of the method of the present invention;

[0072] Figure 4 This is a region map that affects the fast point feature histogram of the sub-point cloud in step S3 of the method of the present invention. Detailed Implementation

[0073] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0074] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0075] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0076] like Figure 1 As shown, this invention provides a point cloud quality assessment method based on relative entropy, comprising the following steps:

[0077] S1: As Figure 2 As shown, the point cloud is resampled through a high-pass filter to obtain the geometric structure.

[0078] A point cloud consists of N points, each point having K attributes. Let's denote this type of point cloud as...

[0079]

[0080] Using geometric distance To calculate the midpoint of the sub-point cloud and The connection weights between them are defined as follows:

[0081]

[0082] The edge density of each vertex is measured using a diagonal matrix D. Connection density The edge density D is defined as follows:

[0083]

[0084] The high-pass filter through which the reference point cloud passes is represented as:

[0085] Where L is the filter length, h l It is the l-th filter coefficient. It is a point cloud shifting operator, defined as:

[0086] A = D -1 W

[0087] Based on h(A), select a Haar-like local domain filter h. HH (A) Implement a high-pass filter, defined as follows:

[0088]

[0089] Where λ i Let U be the eigenvalue of A, and U be the eigenvector of A.

[0090] Each point that passes through the filter The frequency response is:

[0091]

[0092] Let the filter sampling frequency be f. s Total number of points in the sampled point cloud Point cloud after resampling Each point in it is a geometric structure.

[0093] Furthermore, in step S2, given a Euclidean distance threshold γ, the distance between the reference point cloud and the distorted point cloud is... Geometric key points in Constructing a network using the k-nearest neighbor search algorithm centered at the center, such as Figure 3 The sub-point cloud shown is defined as follows;

[0094]

[0095]

[0096] The points in the sub-point cloud are arranged in ascending order of their distance from the geometric key point.

[0097] The specific steps described in S3 for treating relative entropy as a result of the geometric quality of the sub-point cloud include:

[0098] S31: From S r Selecting distance geometric key points Recent k r Points, this k r The maximum distance between a point and the geometric key point is denoted as .

[0099] From S d Select the k closest to the geometric key point. d Points, this k d The maximum distance between a point and the geometric key point is denoted as .

[0100] S32: In S r In the middle, calculation Each pair within the distance and its neighboring points Point feature histogram Figure 3 elements Then, the statistics are compiled into a simple point feature histogram.

[0101] In S d In the middle, calculation Each pair within the distance and its neighboring points Point feature histogram Figure 3 elements Then, the statistics are compiled into a simple point feature histogram.

[0102] S33: In S r In the middle, determine each point exist Find the neighboring points within the distance and calculate each one. Point feature histogram Figure 3 elements Then, the statistics are compiled into a simple point feature histogram.

[0103] In S d In the middle, determine each point exist Histogram of features of points within the distance Figure 3 elements Then, the statistics are compiled into a simple point feature histogram.

[0104] S44: Regions affecting the fast point feature histogram of the first principal component, such as... Figure 4 As shown. Calculate S. r and S d Fast Point Feature Histogram (FPFH) r and FPFH d The definition is as follows:

[0105]

[0106]

[0107] Where d i It is the Euclidean distance between points.

[0108] S35: Calculate FPFH r and FPFH d The relative entropy between them yields the geometric quality assessment index D. i The relative entropy between histograms is defined as:

[0109]

[0110] Furthermore, the specific steps for obtaining the color quality results of the sub-point cloud in S4 are as follows:

[0111] S41: [S] r and S d The color attributes are converted from RGB space to LMN space.

[0112]

[0113] S42: Define the zeroth moment, the first moment, and the second moment as the total number of moments m. g Mean μ g σ g .

[0114] S43: Define S r and S d The three measures of the color attribute are:

[0115]

[0116]

[0117]

[0118] N0, N1, and N3 are three small non-zero numbers to prevent the denominator from being 0.

[0119] S44: Multiply the above three metrics together to get S r and S dThe quality measurement results of the color attributes between them.

[0120]

[0121] Furthermore, the weighted aggregation of the geometric and color quality results of the sub-point clouds in step S5 yields the final tensor point cloud quality assessment model based on relative entropy:

[0122]

[0123] in, and These are the weighting factors for the geometric quality assessment index and the color quality assessment index, respectively.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for point cloud quality assessment based on relative entropy, characterized in that: The method comprises the following steps: S1: down-sampling the reference point cloud, and obtaining the geometric architecture of the down-sampled reference point cloud through a high-pass filter; S2: using the k-nearest neighbor classification algorithm, construct sub-point clouds centered on the geometric architecture in the reference point cloud and the distorted point cloud, respectively denoted as and ; S3: compute and fast point feature histogram and then compute and relative entropy, take relative entropy as the geometric quality result of sub-point cloud; S4: Will and The color attributes are converted from the RGB color space to the LMN color space. The zeroth moment, the first moment, and the second moment of the color space are calculated. The color quality result of the sub-point cloud is obtained by weighted calculation. S5: weighting and aggregating the geometric quality result and the color quality result of the sub point cloud to obtain a final relative-entropy-based point cloud quality evaluation model.

2. The method of claim 1, wherein: The step S1 comprises the following steps: S11: The reference point cloud is composed of N points, each point has M attributes, denoted as: ​​ wherein, represents the 1st to Nth points, represents a space of dimension . S12: Calculate connection weights between points in sub-point cloud using geometric distance where d is the geometric distance between two points.​​ wherein, denotes the variance between points, denotes the Euclidean distance threshold for clustering adjacent points into the same sub-point cloud; S13: Use diagonal matrix to measure the edge density of each vertex, point of connection density , edge density is defined as follows: S14: the reference point cloud passing through the high-pass filter is expressed as: wherein is a filter length, is a first filter coefficient; is a sub-point cloud shift operator defined as: S15: In On the basis of the selected one class Haar local domain filter Realize high-pass filtering, defined as follows: wherein ~ is an eigenvalue of is an eigenvector of S16: each point passing through the filter The frequency response of this filter is: wherein representing a point a set of neighboring points of the point representing a point and a characteristic covariance matrix of the points Let the filter sampling frequency be , the total number of points in the sampled point cloud , and each point in the resampled reference point cloud is a geometric architecture.

3. The method of claim 2, wherein: In step S2, a Euclidean distance threshold is given. In the reference point cloud and the distorted point cloud, the resampled reference point cloud is used. Geometric key points in A sub-point cloud is constructed using the k-nearest neighbor classification algorithm centered at the point, defined as follows: wherein denotes a sub-point cloud centered on the geometric architecture constructed in the reference point cloud, denotes a point in the reference point cloud, denotes the resampled reference point cloud, denotes a sub-point cloud centered on the geometric architecture constructed in the distorted point cloud, denotes a point in the distorted point cloud, denotes the resampled distorted point cloud; The points in the sub point cloud are arranged from small to large according to the distance from the geometric key point.

4. The method of claim 3, wherein: The specific steps of taking the relative entropy as the geometric quality result of the sub point cloud in the step S3 comprise: S31: from the distance geometry key point the nearest point, record the maximum distance of this point from the distance geometry key point as ; From the distance geometry key points closest to the points, the maximum distance of the points from the geometry key points is recorded as ; S32: In In the middle, calculation Each pair within the distance and its neighboring points The three elements of a point feature histogram Then, the statistics are compiled into a simple point feature histogram. ; exist In the middle, calculation Each pair within the distance and its neighboring points The three elements of a point feature histogram Then, the statistics are compiled into a simple point feature histogram. ; S33: In In the middle, determine each point exist Find the neighboring points within the distance and calculate each one. The three elements of a point feature histogram Then, the statistics are compiled into a simple point feature histogram. ; In each point In distance within the point feature histogram three elements , then statistics for a simple point feature histogram ; S34: compute and fast point feature histograms and are defined as follows: wherein is the Euclidean distance between points; S35: Calculate and the relative entropy between the histograms, resulting in a geometric quality assessment indicator The relative entropy between the histograms is defined as: 。 5. The method of claim 4, wherein: The color quality result of the sub point cloud is calculated in the step S4, and the specific steps comprise the following steps: S41: converting the color attributes of the image from the RGB space to the LMN space: and the color attributes of the image from the RGB space to the LMN space: S42: define the zeroth moment, the first moment, and the second moment as the total number , mean , variance ; S43: defining and the 3 metric values of the color properties of: are three small non-zero numbers that prevent the denominator from being zero; S44: multiply the above 3 metric values to get and the quality metric result of color property between 。 6. The method of claim 5, wherein: The weighting and aggregation of the geometric quality result and the color quality result of the sub point cloud in the step S5 to obtain the final relative-entropy-based point cloud quality evaluation model are as follows: wherein, and are weight factors for the geometric quality assessment indicator and the color quality assessment indicator, respectively.

Images