Information Technology Reference
In-Depth Information
Table 2
RMSE of IQMs on LIVE data set
IQM
Blurring
JPEG
JPEG2000
Noise
All
PSNR
9.78
8.43
7.45
2.71
9.60
SSIM
7.50
5.97
5.71
3.89
8.50
MSSIM
5.25
5.43
4.84
4.16
7.11
VSNR
5.94
5.78
5.52
3.35
7.47
VIF
4.39
6.49
5.13
2.97
6.53
UQI
5.09
8.46
8.59
5.53
8.77
IFC
4.99
7.51
7.55
5.50
7.37
NQM
7.55
6.31
6.00
2.79
7.45
WSNR
6.30
6.57
6.97
3.52
7.79
PHVS
6.41
5.81
5.52
2.56
7.71
JND
5.99
6.87
6.11
3.28
8.83
Average
6.29
6.69
6.31
3.66
7.92
Table 3
Pearson correlation coefficients of IQMs on LIVE data set
IQM
Blurring
JPEG
JPEG2000
Noise
All
PSNR
0.783
0.850
0.888
0.986
0.801
SSIM
0.879
0.928
0.936
0.970
0.848
MSSIM
0.943
0.941
0.954
0.943
0.896
VSNR
0.926
0.932
0.940
0.926
0.885
VIF
0.960
0.914
0.949
0.960
0.913
UQI
0.946
0.849
0.848
0.946
0.837
IFC
0.949
0.883
0.885
0.949
0.888
NQM
0.877
0.919
0.929
0.877
0.885
WSNR
0.916
0.912
0.903
0.916
0.874
PHVS
0.913
0.932
0.940
0.913
0.877
JND
0.925
0.901
0.927
0.930
0.835
Average
0.911
0.906
0.918
0.938
0.867
•
The performance of IQMs has significant difference for different distortion
types. For example, PSNR is suitable for evaluating the quality degradation
caused by noise, while its performance on blurring distortion is not promising.
UQI and IFC have excellent performance in predicting the compression
degradation.
•
Statistically speaking, these IQMs have promising performance on a single dis-
tortion type, while the robustness to the change of the distortion types is not
very strong.
