Table 5.

Results of different IQA algorithms for the PIPAL dataset. The upward arrows indicate that higher values are better for this metric and vice versa, with the best results marked in bold.

Method	Traditional Method	PSNR-Oriented Method	GAN-Based Method	All Images Recovery Method
PSNR	0.4782	0.5462	0.2839	0.4099
$\mathrm{NQM}\uparrow$ [37]	0.5374	0.6462	0.3410	0.4742
$\mathrm{UQI}\uparrow$ [38]	0.6087	0.7060	0.3385	0.5257
$\mathrm{SSIM}\uparrow$ [1]	0.5856	0.6897	0.3388	0.5209
$\mathrm{MS}-\mathrm{SSIM}\uparrow$ [9]	0.6527	0.7528	0.3823	0.5596
$\mathrm{IFC}\uparrow$ [16]	0.7062	0.8244	0.3217	0.5651
$\mathrm{VIF}\uparrow$ [17]	0.6927	0.7864	0.3857	0.5917
$\mathrm{VSNR}-\mathrm{FR}\uparrow$ [39]	0.6146	0.7076	0.3128	0.5086
$\mathrm{RFSIM}\uparrow$ [40]	0.4593	0.5525	0.2951	0.4232
$\mathrm{GSM}\uparrow$ [10]	0.6074	0.6904	0.3523	0.5361
$\mathrm{SR}-\mathrm{SIM}\uparrow$ [41]	0.6561	0.7476	0.4631	0.6094
$\mathrm{FSIM}\uparrow$ [20]	0.6515	0.7381	0.4090	0.5896
${\mathrm{FSIM}}_{\mathrm{c}}\uparrow$ [20]	0.6509	0.7374	0.4058	0.5872
$\mathrm{VSI}\uparrow$ [42]	0.6086	0.6938	0.3706	0.5475
$\mathrm{MAD}\downarrow$ [43]	0.6720	0.7575	0.3494	0.5424
$\mathrm{RDIE}\downarrow$	0.6476	0.7203	0.5280	0.6368