236
X. Qi et al.
Fig. 2. The structure of generator of MCGAN.
T1
T2
T3
T4
T5
Fig. 3. The images of ground truth and those derived by CGAN, NW-MCGAN, MCGAN in
multi-period. The periods are T1 (18 00, 22:00), T2 (22 00, 3:00), T3 (3 00, 8:00), T4 (8 00,
13:00) and T5 (13 00, 18:00]) separately.
Table 1. Mean structural similarity of the original and derived images
Period
T1
T2
T3
T4
T5
Method
Index
CGAN
l
0.9998 0.9898 0.9931 0.8534 0.9840
c
0.9957 0.9910 0.9629 0.8322 0.8850
s
0.9860 0.9745 0.9561 0.8481 0.8387
MSSIM 0.9821 0.9567 0.9187 0.6063 0.7395
NW-MCGAN l
0.9993 0.9949 0.9962 0.9786 0.9901
c
0.9925 0.9915 0.9787 0.9573 0.8938
(continued)
X. Qi et al.
Fig. 2. The structure of generator of MCGAN.
T1
T2
T3
T4
T5
Fig. 3. The images of ground truth and those derived by CGAN, NW-MCGAN, MCGAN in
multi-period. The periods are T1 (18 00, 22:00), T2 (22 00, 3:00), T3 (3 00, 8:00), T4 (8 00,
13:00) and T5 (13 00, 18:00]) separately.
Table 1. Mean structural similarity of the original and derived images
Period
T1
T2
T3
T4
T5
Method
Index
CGAN
l
0.9998 0.9898 0.9931 0.8534 0.9840
c
0.9957 0.9910 0.9629 0.8322 0.8850
s
0.9860 0.9745 0.9561 0.8481 0.8387
MSSIM 0.9821 0.9567 0.9187 0.6063 0.7395
NW-MCGAN l
0.9993 0.9949 0.9962 0.9786 0.9901
c
0.9925 0.9915 0.9787 0.9573 0.8938
(continued)