No-reference point cloud quality assessment based on fusion of 3D and 2D features

Type	Metric	CPCD2.0				SJTU-PCQA				IRPC
Type	Metric	PLCC	SROCC	KROCC	RMSE	PLCC	SROCC	KROCC	RMSE	PLCC	SROCC	KROCC	RMSE
Full-reference	P2point_MSE^[3]	0.6784	0.5491	0.4142	0.8617	0.4721	0.4096	0.2857	2.1394	0.3357	0.3281	0.2146	0.9313
	P2plane_Hausdorff^[4]	0.4061	0.3786	0.2663	1.0718	0.3752	0.4609	0.3354	2.4467	0.3925	0.2541	0.1975	0.9089
	P2plane_MSE^[4]	0.6914	0.5692	0.4385	0.8474	0.5651	0.4956	0.3514	2.0022	0.4296	0.2564	0.1957	0.9089
	PC-MSDM^[10]	0.6254	0.5321	0.3842	0.9152	0.4123	0.3241	0.2189	2.2110	0.2729	0.1519	0.1063	0.9515
	PointSSIM-G^[13]	0.5343	0.5533	0.4238	0.9915	0.3860	0.3649	0.2792	2.2410	0.6183	0.5951	0.4693	0.7760
	PointSSIM-C^[13]	0.7457	0.6891	0.4863	0.7814	0.4561	0.4185	0.3172	2.1598	0.6648	0.5638	0.4211	0.7376
	PCQM^[11]	0.4813	0.3408	0.2615	1.0281	0.7771	0.7420	0.5624	1.5274	0.5611	0.5611	0.3033	0.8184
	GraphSIM^[12]	0.8553	0.8296	0.6234	0.6077	0.8900	0.8800	-	1.1300	0.9400	0.7600	-	0.2100
No-reference	BEQ-CVP^[14]	0.7950	0.7890	0.5983	0.7218	0.9192	0.8972	0.7343	0.9717	0.7265	0.7298	0.5427	0.6586
	IW-SSIM^[15]	-	-	-	-	0.7949	0.7833	-	-	0.0911	0.1339	-	-
	MFPCQA^[16]	-	-	-	-	0.8972	0.8894	-	0.6488	-	-	-	-
	Proposed	0.9203	0.8996	0.7494	0.4195	0.9463	0.9248	0.8231	0.3854	0.9125	0.8566	0.7018	0.4529

Table 2. Comparison of overall performance of different methods on different datasets

View table

View in Article

Table 2. Comparison of overall performance of different methods on different datasets

Type	Metric	CPCD2.0 subset				ICIP2020				M-PCCD
Type	Metric	PLCC	SROCC	KROCC	RMSE	PLCC	SROCC	KROCC	RMSE	PLCC	SROCC	KROCC	RMSE
No-reference	PRPCQA^[17]	0.8591	0.8351	0.7014	0.5581	0.9310	0.9307	0.8814	0.3011	0.9144	0.9241	0.6857	0.5107
	VPPCQA^[21]	0.8343	0.8460	0.6578	0.6046	0.9114	0.9264	0.8311	0.3965	0.9147	0.9322	0.6923	0.4903
	Proposed	0.9100	0.8842	0.7348	0.4371	0.9160	0.8827	0.8564	0.3672	0.9215	0.9400	0.7638	0.4720

Table 3. Performance comparison of different features on CPCD2.0 dataset
View table
View in Article
Table 3. Performance comparison of different features on CPCD2.0 dataset
$f_{2 D}$ $f_{3 D}$ PLCC SROCC KROCC RMSE
√ 0.9096 0.8879 0.6792 0.4391
√ 0.5148 0.4754 0.5148 0.9596
√ √ 0.9203 0.8996 0.7494 0.4195

Table 4. Performance comparison of different feature fusion methods
View table
View in Article
Table 4. Performance comparison of different feature fusion methods
Method PLCC SROCC KROCC RMSE
Add 0.8985 0.8757 0.7163 0.4626
Concat 0.9141 0.8914 0.7370 0.4236
SCMA 0.9203 0.8996 0.7494 0.4195

Table 5. The impact of the number of point cloud sub-models and their inclusion points on the performance of the proposed method
View table
View in Article
Table 5. The impact of the number of point cloud sub-models and their inclusion points on the performance of the proposed method
Parameter PLCC SROCC KROCC RMSE
48×256 0.9060 0.8847 0.7327 0.4495
24×512 0.9049 0.8902 0.7249 0.4557
12×1024 0.8821 0.8607 0.6949 0.4993
6×2048 0.9203 0.8996 0.7494 0.4195

Table 6. Performance comparison of different center point generation methods for point cloud sub-models
View table
View in Article
Table 6. Performance comparison of different center point generation methods for point cloud sub-models
Sampling method PLCC SROCC KROCC RMSE
Random sampling 0.8427 0.8236 0.7129 0.4572
Cell sampling 0.8668 0.7707 0.7018 0.4327
Geometric sampling 0.8643 0.8695 0.6821 0.4311
Farthest point sampling 0.9203 0.8996 0.7494 0.4195

Table 7. Performance comparison of the number of different cloud points' patch
View table
View in Article
Table 7. Performance comparison of the number of different cloud points' patch
Patch number PLCC SROCC KROCC RMSE
2 0.8321 0.8014 0.6929 0.4723
4 0.8718 0.8207 0.7381 0.4532
6 0.9203 0.8996 0.7494 0.4195
8 0.9000 0.8712 0.7301 0.4330

Tools

Get Citation

Copy Citation Text

Taiwei Liu, Mei Yu, Renwei Tu. No-reference point cloud quality assessment based on fusion of 3D and 2D features[J]. Opto-Electronic Engineering, 2025, 52(4): 250001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Article

Received: Jan. 1, 2025

Accepted: Feb. 28, 2025

Published Online: Jun. 11, 2025

The Author Email: Mei Yu (郁梅)

DOI:10.12086/oee.2025.250001

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Comparison of overall performance of different methods on different datasets

Table 1. Comparison of overall performance of different methods on different datasets

Table 2. Comparison of overall performance of different methods on different datasets

Table 2. Comparison of overall performance of different methods on different datasets

Table 3. Performance comparison of different features on CPCD2.0 dataset

Table 3. Performance comparison of different features on CPCD2.0 dataset

Table 4. Performance comparison of different feature fusion methods

Table 4. Performance comparison of different feature fusion methods

Table 5. The impact of the number of point cloud sub-models and their inclusion points on the performance of the proposed method

Table 5. The impact of the number of point cloud sub-models and their inclusion points on the performance of the proposed method

Table 6. Performance comparison of different center point generation methods for point cloud sub-models

Table 6. Performance comparison of different center point generation methods for point cloud sub-models

Table 7. Performance comparison of the number of different cloud points' patch

Table 7. Performance comparison of the number of different cloud points' patch