A perceptual quality assessment method for V-PCC coded point clouds

Abstract

The present invention belongs to the technical field of point cloud quality assessment, and relates to a V‑PCC coded point cloud perception quality assessment method. First, a V‑PCC coded point cloud bit stream is simply parsed to obtain texture quantization parameters, texture bit rate, and geometric quantization parameters. Then, the texture complexity of the point cloud is predicted according to the texture quantization parameters and the texture bit rate. Then, a point cloud texture distortion assessment model is trained in the geometric lossless state in combination with the texture quantization parameter input to predict the point cloud quality in the geometric lossless state. The attenuation of the geometric distortion is calculated according to the geometric quantization parameters. Finally, the texture distortion and the geometric distortion are fused to obtain a predicted value of the distorted point cloud quality. The method can accurately assess the quality of 3D point clouds, has low computational complexity, and high real-time performance. The method can be applied to any point cloud business-related field, has universality, and can predict a quality score that takes into account the masking effect of the human eye. The quality score of the distorted point cloud is more accurately assessed, and the objective quality score has better consistency with the subjective quality score.

Description

Technical Field

[0001] The present invention belongs to the technical field of point cloud quality assessment and relates to a V-PCC coded point cloud perception quality assessment method. Background Art

[0002] 3D point clouds are increasingly used in manufacturing, construction, environmental monitoring, autonomous driving, animation, and other fields. Usually, point cloud applications require the best effort delivery network to transmit compressed point clouds, but the quality of the received point clouds cannot be guaranteed. Therefore, developing an objective point cloud quality assessment model is necessary for system design, experience quality prediction, and quality monitoring.

[0003] According to the different input information, point cloud quality assessment can be divided into parameter models, packet layer models, bitstream layer models, media layer models and hybrid models. The parameter model uses parameters from the network or application such as packet loss rate, delay, coding bit rate, frame rate, etc. to predict the point cloud quality. These models were originally designed to roughly estimate the end-to-end service quality; the packet layer model analyzes the packet header to obtain information about the service quality. In addition to the parameters used in the parameterized model, the packet layer model can also use more information to predict the point cloud quality, including frame type, frame bit rate, bit rate ratio of intra-frame and inter-frame coded frames, and the location of the lost data packet in the network, so the packet layer model is very suitable for non-intrusive quality monitoring; in addition to the packet layer model data, the bitstream layer model can also use the media-related payload information in the compressed code stream. This model is used to design non-intrusive monitoring and emphasizes accuracy; the media layer model, also known as the voxel-based model, uses the reconstructed point cloud signal to predict the point cloud quality; the hybrid model can use information from the bitstream layer and the voxel domain to predict the point cloud quality.

[0004] Generally speaking, getting more information from the bitstream means an increase in the computational complexity of the model. For point cloud quality assessment, it is very important to find a compromise between accuracy and computational complexity. The parameter model and the packet model have poor accuracy because they cannot access the payload information. Although the hybrid model and the media layer model have high accuracy, their computation is complex because they need to decode to obtain the media layer information. Moreover, the pure media layer model cannot obtain the packet layer or bitstream layer information, but the packet layer or bitstream layer information is very important for point cloud quality assessment. Therefore, it is necessary to design a V-PCC coded point cloud perceptual quality assessment method using a bitstream layer model to provide higher accuracy while reducing computational complexity. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and design a V-PCC coded point cloud perceptual quality assessment method. The adopted bitstream layer model takes into account the masking effect of visual signals on the human eye, so that the objective quality prediction value has good consistency with the subjective quality score and has real-time performance, so as to guide various processing processes such as 3D point cloud encoding and transmission.

[0006] In order to achieve the above object, the specific process of V-PCC coded point cloud perceptual quality assessment of the present invention includes the following steps:

[0007] S1. Perform V-PCC encoding on the original point cloud, select different degrees of geometric distortion and texture distortion combinations, obtain a 3D point cloud database with encoding distortion, and conduct subjective evaluation experiments to obtain the subjective quality score of each distorted point cloud;

[0008] S2, parsing the V-PCC coded point cloud code stream to obtain geometric quantization parameters, texture quantization parameters and texture bit rate;

[0009] S3. Establish a 3D point cloud texture complexity prediction model. Under different texture quantization parameters, the texture complexity and texture bit rate both show an approximately linear relationship. Based on the 3D point cloud texture complexity prediction model, each encoded point cloud in the 3D point cloud training set is bitstream parsed to achieve prediction calculation of 3D point cloud texture complexity.

[0010] S4, according to the texture complexity obtained in S3 and the texture quantization parameter obtained in S2, the texture complexity, texture quantization parameter and subjective evaluation score of the distorted point cloud are used as training data to fit the model parameters, and a geometric lossless point cloud texture distortion assessment model is obtained;

[0011] S5. Using the distorted point cloud training set, perform logistic curve fitting on the point cloud subjective evaluation score and geometric quantization parameter to obtain the unknown parameters of the attenuation model, obtain the 3D point cloud geometric distortion model, and realize the mapping of the distorted point cloud geometric quantization parameter GQP to the point cloud subjective evaluation score;

[0012] S6. The point cloud texture distortion prediction value obtained according to the geometric lossless point cloud texture distortion assessment model in S4 is fused with the geometric attenuation value obtained according to the 3D point cloud geometric distortion model in S5 to obtain a fusion model, and the prediction value of the distorted point cloud quality is calculated according to the fusion model to realize the bitstream layer evaluation of the perceptual quality of the V-PCC encoded point cloud.

[0013] The 3D point cloud texture complexity prediction model established in S3 of the present invention is:

[0014]

[0015] Where TC is the texture complexity prediction value of the 3D point cloud, TQP is the texture quantization parameter, TBPP is the texture bit rate, and All follow Changes are calculated as follows:

[0016]

[0017]

[0018] in , , , , , are unknown coefficients obtained by least squares fitting.

[0019] The geometric lossless point cloud texture distortion assessment model established in S4 of the present invention is:

[0020]

[0021] in is the predicted value of the objective quality score of the point cloud texture distortion, Quantize the texture step size: , and is the unknown coefficient, and α and TC are approximately linearly related: , c and d are unknown coefficients. All unknown coefficients of this model are obtained by fitting based on the 3D point cloud training set.

[0022] The 3D point cloud geometric distortion model constructed in S5 of the present invention is:

[0023]

[0024] in Indicates the degree of attenuation of point cloud quality with the geometric quantization parameter GQP, , , , are unknown coefficients obtained by least squares fitting.

[0025] The fusion model described in S6 of the present invention is:

[0026]

[0027] Among them, MOS est It is the final objective quality prediction value of the 3D point cloud.

[0028] The present invention first performs a simple analysis on the V-PCC encoded point cloud bit stream to obtain texture quantization parameters, texture bit rate, and geometric quantization parameters, and then predicts the texture complexity of the point cloud according to the texture quantization parameters and texture bit rate. Then, a point cloud texture distortion assessment model is trained in the geometric lossless state in combination with the texture quantization parameter input to predict the point cloud quality in the geometric lossless state, and the attenuation of the geometric distortion is calculated according to the geometric quantization parameters. Finally, the texture distortion and geometric distortion are fused to obtain the predicted value of the distorted point cloud quality.

[0029] Compared with the prior art, the present invention has the following advantages: first, it can accurately evaluate the quality of 3D point clouds, and then guide the processing process of 3D point clouds, so as to improve the performance of subsequent applications; second, the computational complexity is low, the transmission bandwidth required for quality evaluation in the network is small, the real-time performance is high, and it can be applied to any point cloud business-related fields, and it has universal applicability; third, the texture complexity prediction model and objective quality evaluation model obtained through training can predict the quality score taking into account the masking effect of the human eye, and more accurately evaluate the quality score of distorted point clouds, and the objective quality score has better consistency with the subjective quality score. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 The specific process flow chart of the perceptual quality assessment of V-PCC coded point cloud is obtained by the present invention.

[0031] Figure 2 are the original point cloud and partially distorted point cloud used in the test set in the embodiment of the present invention, where Figure 2 (a), (c), and (e) are the original point clouds of biscuits, coffee_cup, and croissant, respectively; Figure 2 (b), (d), and (f) are the point clouds with different V-PCC encoding parameters corresponding to the above original point clouds: (b) GQP=44, TQP=44, MOS=56.95; (d) GQP=38, TQP=38, MOS=65.28; (f) GQP=26, TQP=26, MOS=89.81. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] Example:

[0034] The specific process of V-PCC coded point cloud perceptual quality assessment in this embodiment is as follows: Figure 1 As shown, the following steps are included:

[0035] S1. Perform V-PCC encoding on the original point cloud, select different degrees of geometric distortion and texture distortion combinations, obtain a 3D point cloud database with encoding distortion, and conduct subjective evaluation experiments to obtain the subjective quality score of each distorted point cloud;

[0036] S2, parsing the V-PCC coded point cloud code stream to obtain geometric quantization parameters, texture quantization parameters and texture bit rate;

[0037] S3. Establish a 3D point cloud texture complexity prediction model. When the texture quantization parameter TQP is different, the texture complexity TC and the texture bit rate TBPP both show an approximately linear relationship. Based on the 3D point cloud texture complexity prediction model, each coded point cloud in the 3D point cloud training set is bitstream parsed to achieve the prediction calculation of the 3D point cloud texture complexity. The established 3D point cloud texture complexity prediction model is:

[0038]

[0039] Where TC is the texture complexity prediction value of the 3D point cloud, TQP is the texture quantization parameter, TBPP is the texture bit rate, and All follow Changes are calculated as follows:

[0040]

[0041]

[0042] in , , , , , are unknown coefficients, obtained by least squares fitting;

[0043] S4. According to the texture complexity obtained in S3 and the texture quantization parameter obtained in S2, the texture complexity, texture quantization parameter and subjective evaluation score of the distorted point cloud are used as training data to fit the model parameters, and the geometric lossless point cloud texture distortion evaluation model is obtained: ,in is the predicted value of the objective quality score of the point cloud texture distortion, Quantize the texture step size: , and is the unknown coefficient, and α and TC are approximately linearly related: , c and d are unknown coefficients. All unknown coefficients of this model are obtained by fitting based on the 3D point cloud training set;

[0044] S5. Using the distorted point cloud training set, the subjective evaluation score of the point cloud and the geometric quantization parameter are fitted with a Logistic curve to obtain the unknown parameters of the attenuation model, and the 3D point cloud geometric distortion model is obtained to achieve the mapping of the distorted point cloud geometric quantization parameter GQP to the subjective evaluation score of the point cloud; the constructed 3D point cloud geometric distortion model is:

[0045]

[0046] in Indicates the degree of attenuation of point cloud quality with the geometric quantization parameter GQP, , , , are unknown coefficients, obtained by least squares fitting;

[0047] S6, the point cloud texture distortion prediction value obtained according to the geometric lossless point cloud texture distortion assessment model in S4 is fused with the geometric attenuation value obtained according to the 3D point cloud geometric distortion model in S5 to obtain a fusion model, and the prediction value of the distorted point cloud quality is calculated according to the fusion model to achieve the bitstream layer evaluation of the perceptual quality of the V-PCC coded point cloud, wherein the fusion model is:

[0048]

[0049] Among them, MOS est It is the final objective quality prediction value of the 3D point cloud.

[0050] In order to evaluate the performance of the objective quality evaluation model, this embodiment uses the following measurement methods to calculate the consistency between the objective quality score and the subjective quality score for the biscuits, cake, cauliflower, coffee_cup and croissant cases: Spearman Rank-order Correlation Coefficient (SRCC), Pearson Linear Correlation Coefficient (PLCC) and Root Mean Square Error (RMSE). The results are shown in the figure. Figure 2 As shown in Table 1, Table 1 fully illustrates the consistency between the objective score and the subjective score obtained for the test point cloud using the quality assessment model of this embodiment.

[0051] Table 1: Performance of the quality assessment model in this embodiment

[0052] Point Cloud SRCC PLCC RMSE biscuits 0.9831 0.9834 2.7724 coffee_cup 0.9092 0.9606 2.8798 croissant 0.9669 0.9614 5.5661

[0053] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for perceptual quality assessment of V-PCC coded point clouds, characterized in that: The following steps are involved: S1. Perform V-PCC encoding on the original point cloud, select different degrees of geometric distortion and texture distortion combinations, obtain a 3D point cloud database with encoding distortion, and conduct subjective evaluation experiments to obtain the subjective quality score of each distorted point cloud; S2, parsing the V-PCC coded point cloud code stream to obtain geometric quantization parameters, texture quantization parameters and texture bit rate; S3. Establish a 3D point cloud texture complexity prediction model. Under different texture quantization parameters, the texture complexity and texture bit rate both show an approximately linear relationship. Based on the 3D point cloud texture complexity prediction model, each encoded point cloud in the 3D point cloud training set is bitstream parsed to achieve prediction calculation of 3D point cloud texture complexity. S4, according to the texture complexity obtained in S3 and the texture quantization parameter obtained in S2, the texture complexity, texture quantization parameter and subjective evaluation score of the distorted point cloud are used as training data to fit the model parameters, and a geometric lossless point cloud texture distortion assessment model is obtained; S5. Using the distorted point cloud training set, perform logistic curve fitting on the point cloud subjective evaluation score and geometric quantization parameters to obtain the unknown parameters of the attenuation model, obtain the 3D point cloud geometric distortion model, and realize the mapping of the distorted point cloud geometric quantization parameters to the point cloud subjective evaluation score; S6. The point cloud texture distortion prediction value obtained according to the geometric lossless point cloud texture distortion assessment model in S4 is fused with the geometric attenuation value obtained according to the 3D point cloud geometric distortion model in S5 to obtain a fusion model, and the prediction value of the distorted point cloud quality is calculated according to the fusion model to realize the bitstream layer evaluation of the perceptual quality of the V-PCC encoded point cloud.

2. The V-PCC coded point cloud perceptual quality assessment method according to claim 1, characterized in that: The 3D point cloud texture complexity prediction model established in S3 is: Where TC is the texture complexity prediction value of the 3D point cloud, TQP is the texture quantization parameter, TBPP is the texture bit rate, and All follow Changes are calculated as follows: in , , , , , are unknown coefficients obtained by least squares fitting.

3. The V-PCC coded point cloud perceptual quality assessment method according to claim 2, characterized in that: The geometric lossless point cloud texture distortion evaluation model established in S4 is: in is the predicted value of the objective quality score of the point cloud texture distortion, Quantize the texture step size: , and is the unknown coefficient, and α and TC are approximately linearly related: , c and d are unknown coefficients. All unknown coefficients of this model are obtained by fitting based on the 3D point cloud training set.

4. The V-PCC coded point cloud perceptual quality assessment method according to claim 3, characterized in that: The 3D point cloud geometric distortion model constructed in S5 is: in Indicates the degree of attenuation of point cloud quality with the geometric quantization parameter GQP, , , , are unknown coefficients obtained by least squares fitting.

5. The V-PCC coded point cloud perceptual quality assessment method according to claim 4, characterized in that: The fusion model in S6 is: Among them, MOS est It is the final objective quality prediction value of the 3D point cloud.

Images