Depth Image Super-Resolution Reconstruction Network Based on Dual Feature Fusion Guidance

Fig. 7. Super-resolution results of Art depth images processed by different algorithms(4×). (a) Original drawings; (b) partial true picture; (c) TGV algorithm; (d) RMRF algorithm; (e) GF algorithm; (f) JBU algorithm; (g) MSG-Net algorithm; (h) FDKN algorithm; (i) RCAN algorithm; (g) DepthSRNet algorithm; (k) proposed algorithm

Download full size

Fig. 8. Super-resolution results of Laundry depth images processed by different algorithms(4×). (a) Original drawings; (b) Partial true picture; (c) TGV algorithm; (d) RMRF algorithm; (e) GF algorithm; (f) JBU algorithm; (g) MSG-Net algorithm; (h) FDKN algorithm; (i) RCAN algorithm; (g) Depth-SR algorithm; (k) proposed algorithm

Download full size

Table 1. RMSE/PSNR(dB) index results after super-resolution reconstruction

View table

Table 1. RMSE/PSNR(dB) index results after super-resolution reconstruction

Method	Scale	Art	Books	Dolls	Laundry	Moebius	Reindeer	Average
TGV^［3］	2×	3.16/38.14	1.33/45.62	1.17/46.77	1.87/42.69	1.14/46.99	2.40/40.53	1.85/43.46
	4×	3.73/36.70	1.67/43.68	1.42/45.09	2.25/41.09	1.45/44.90	2.65/39.67	2.20/41.86
	8×	7.12/31.08	2.27/41.01	2.05/41.90	4.05/35.98	2.41/40.49	4.33/35.40	3.71/37.64
	16×	12.08/26.49	4.89/34.34	4.44/35.18	8.01/30.06	5.41/33.47	9.05/29.00	7.31/31.42
RMRF^［4］	2×	2.31/ 40.85	1.24/46.24	1.24/46.29	1.55/44.31	1.23/46.34	1.80/43.01	1.56/44.51
	4×	3.50/ 37.25	1.81/42.96	1.64/43.84	2.32/40.82	1.74/43.30	2.62/39.76	2.27/41.32
	8×	4.79/ 34.52	2.46/40.32	2.15/41.47	3.05/38.43	2.49/40.20	3.24/37.93	3.03/38.81
	16×	7.18/ 31.01	3.32/37.71	2.97/38.67	4.46/35.14	3.28/37.80	4.51/35.03	4.29/35.89
GF^［14］	2×	3.15/38.16	1.43/44.99	1.22/46.40	2.01/42.09	1.23/46.33	2.25/41.11	1.88/43.18
	4×	3.90/36.30	1.76/43.21	1.64/43.83	2.47/40.29	1.73/43.37	2.82/39.14	2.38/41.02
	8×	5.43/33.44	2.38/40.59	2.26/41.05	3.44/37.41	2.46/40.31	3.95/36.20	3.49/38.17
	16×	8.15/29.90	3.34/37.65	3.45/37.37	5.03/34.10	4.16/35.75	5.83/32.81	4.99/34.60
JBU^［15］	2×	2.69/39.53	1.11/47.22	1.02/47.96	1.60/44.05	0.91/48.95	1.87/42.69	1.53/45.07
	4×	4.04/36.00	1.88/42.65	1.49/44.67	2.64/39.70	1.52/44.49	2.86/39.00	2.41/41.09
	8×	5.23/33.76	2.49/40.21	1.85/42.79	3.44/37.40	2.18/41.36	3.60/37.00	3.13/38.75
	16×	7.13/31.07	3.96/36.18	2.52/40.10	5.96/32.63	3.08/38.36	4.38/35.30	4.51/35.61
MSG-Net^［6］	2×	0.66/51.74	0.37/56.77	0.37/56.77	0.79/50.18	0.31/58.30	0.42/55.67	0.49/54.91
	4×	1.47/44.78	0.67/51.61	0.73/50.86	0.79/50.18	0.58/52.86	0.98/48.31	0.87/49.77
	8×	2.46/40.31	1.03/47.87	1.10/47.30	1.51/44.55	0.94/48.67	1.76/43.22	1.47/45.32
	16×	4.57/34.93	1.63/43.88	1.63/43.89	2.63/39.73	1.69/43.57	2.92/38.79	2.51/40.80
FDKN^［16］	2×	1.53/44.44	0.50/54.15	0.70/51.23	0.88/49.24	0.60/52.57	1.09/47.38	0.88/49.84
	4×	2.10/41.67	0.73/50.86	0.93/48.76	1.26/46.12	0.79/50.18	1.50/44.61	1.22/47.03
	8×	3.16/38.14	1.21/46.48	1.30/45.85	2.00/42.11	1.24/46.26	2.27/41.01	1.86/43.31
	16×	9.46/28.61	3.93/36.24	3.21/38.00	5.95/32.64	4.26/35.54	6.53/31.83	5.56/33.81
RCAN^［17］	2×	1.01/48.04	0.37/56.77	0.47/54.69	0.55/53.32	0.43/55.46	0.72/50.98	0.57/53.21
	4×	1.51/44.55	0.56/53.17	0.70/51.23	0.86/49.44	0.60/52.57	1.08/47.46	0.89/49.74
	8×	2.20/41.28	0.85/49.54	1.02/47.96	1.33/45.65	0.88/49.24	1.57/44.21	1.31/46.31
	16×
Depth-SRNet^［7］	2×	0.53/53.63	0.42/55.66	0.49/54.33	0.44/55.25	0.44/55.26	0.51/53.88	0.47/54.67
	4×	1.20/46.55	0.60/52.49	0.81/49.96	0.78/50.26	0.68/51.48	0.96/48.51	0.84/49.88
	8×	2.22/41.22	0.89/49.12	1.11/47.22	1.31/45.81	0.96/48.49	1.57/44.21	1.34/46.01
	16×	3.90/36.30	1.51/44.54	1.54/44.38	2.26/41.06	1.56/44.27	2.47/40.29	2.21/41.81
proposed method	2×	0.41/55.87	0.30/58.59	0.35/57.25	0.33/57.76	0.32/58.03	0.38/56.54	0.35/57.34
	4×	0.95/48.57	0.42/55.67	0.61/52.42	0.59/52.71	0.50/54.15	0.79/50.18	0.64/52.28
	8×	2.15/41.48	0.76/50.51	0.97/48.39	1.15/46.91	0.73/50.86	1.48/44.73	1.21/47.15
	16×	3.72/36.72	1.45/44.90	1.45/44.90	2.25/41.09	1.34/45.59	2.32/40.82	2.09/42.34

Table 2. Results of ablation experiment
View table
Table 2. Results of ablation experiment
Network RMSE PSNR /dB
M1 1.11 47.23
M2 1.07 47.54
M3 0.99 48.23
M4 0.95 48.57

Tools

Get Citation

Copy Citation Text

Haowen Geng, Yu Wang, Yanling Xin. Depth Image Super-Resolution Reconstruction Network Based on Dual Feature Fusion Guidance[J]. Laser & Optoelectronics Progress, 2024, 61(8): 0837007

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites