Visual Perception Evaluation Method of Stereo Images Based on CNN-SVR

Fig. 2. Acquisition process of potential significant discomfort area. (a) Left view; (b) right view; (c) disparity map; (d) planar saliency map; (e) stereoscopic saliency map; (f) salient discomfort region

Download full size

Fig. 3. Part of the reconstructed images

Download full size

Fig. 4. Disparity map of potential significant discomfort area

Download full size

Fig. 5. Regression prediction process of CNN-SVR

Download full size

Fig. 6. Part images in the public stereo image databases

Download full size

Fig. 7. Scatter plot of the relationship between model-predicted value and subjective evaluation value

Download full size

Table 1. Performance comparison of different models on the NBU S3D-VCA dataset

View table

Table 1. Performance comparison of different models on the NBU S3D-VCA dataset

Model	PLCC	SROCC	RMSE
Model in Ref.［8］	0.8128	0.7684	0.4015
Model in Ref.［28］	0.8240	0.7870	0.5580
Model in Ref.［2］	0.8310	0.7980	0.2780
Model in Ref.［29］	0.8358	0.8131	0.4973
Model in Ref.［3］	0.8668	0.8034	0.4732
Proposed model	0.8759	0.8335	0.4010

Table 2. Performance comparison of different models on the IVY LAB dataset

View table

Table 2. Performance comparison of different models on the IVY LAB dataset

Model	PLCC	SROCC	RMSE
Model in Ref.［30］	0.7361	0.7190	0.4704
Model in Ref.［2］	0.8290	0.8300	0.2350
Model in Ref.［31］	0.8358	0.8131	0.4973
Model in Ref.［3］	0.9315	0.8976	0.2641
Proposed model	0.9469	0.9035	0.2547

Table 3. Performance comparison of different regression methods
View table
Table 3. Performance comparison of different regression methods
Regression method NBU S3D-VCA IVY LAB
PLCC SROCC RMSE PLCC SROCC RMSE
CNN 0.7886 0.7284 0.5973 0.8336 0.8037 0.4375
SVR 0.8617 0.8162 0.5428 0.8945 0.8672 0.3560
CNN-SVR 0.8759 0.8335 0.4010 0.9469 0.9035 0.2547

Table 4. Comparison of different feature extraction methods
View table
Table 4. Comparison of different feature extraction methods
Method PLCC SROCC RMSE
Global feature 0.7271 0.6307 0.5994
Local feature 0.8173 0.7614 0.5042
Proposed method 0.8759 0.8335 0.4010

Tools

Get Citation

Copy Citation Text

Yuan Xu, Chunyi Chen, Xiaojuan Hu, Haiyang Yu, Ye Tian. Visual Perception Evaluation Method of Stereo Images Based on CNN-SVR[J]. Laser & Optoelectronics Progress, 2023, 60(8): 0811027

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites