Table 3 Quantitative evaluation results of virtual restoration of regions 1–2

From: Virtual restoration of ancient mold-damaged paintings based on spectral-guided asymmetric autoencoder for hyperspectral images

Region

RMSE

MAPE

MAE

RMSE

MAPE

MAE

Region 1

White material on the hand

Brown background

Original

0.040

7.109

0.031

0.032

6.502

0.026

Inverse PCA

0.031

5.462

0.024

0.039

9.503

0.035

Inverse MNF

0.030

5.555

0.024

0.026

5.583

0.022

3D CNN

0.037

6.012

0.027

0.040

7.755

0.031

U-Net

0.032

5.665

0.024

0.038

9.186

0.033

Proposed

0.027

4.475

0.020

0.022

4.043

0.017

 

Dark gray clothing background

Light gray clothing background

Original

0.032

8.098

0.027

0.077

14.456

0.063

Inverse PCA

0.027

6.240

0.024

0.064

12.652

0.055

Inverse MNF

0.036

9.183

0.034

0.058

11.480

0.048

3D CNN

0.027

6.703

0.023

0.056

9.660

0.043

U-Net

0.031

6.795

0.026

0.065

13.094

0.057

Proposed

0.022

5.190

0.018

0.049

7.373

0.033

Region 2

Brown paper background

   

Original

0.051

11.360

0.043

   

Inverse PCA

0.051

11.389

0.043

   

Inverse MNF

0.031

6.261

0.025

   

3D CNN

0.028

5.309

0.022

   

U-Net

0.036

7.338

0.029

   

Proposed

0.030

5.967

0.024

   
  1. Optimal results are displayed in bold, while suboptimal results are underlined.