Double Branch Residual Network for Demosaicing

Jihui Yu and Xiaomin Yang^*

Author Affiliations

College of Electronic Information, Sichuan University, Chengdu, Sichuan 610065, China

show less

DOI: 10.3788/LOP202158.0210012 Cite this Article Set citation alerts

Jihui Yu, Xiaomin Yang. Double Branch Residual Network for Demosaicing[J]. Laser & Optoelectronics Progress, 2021, 58(2): 0210012 Copy Citation Text

show less

Fig. 1. Bayer-type CFA. (a) Basic structure image; (b) CFA image

Fig. 2. Green channel process

Fig. 3. Overall network structure

Fig. 4. Residual attention model

Fig. 5. Structure of embed residual module

Fig. 6. Subjective results

Channel	AP	AHD	DLMMSE	GBTF	LSSC
Channel	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM
R	38.00/0.9790	36.98/0.9718	39.17/0.979	39.66/0.9834	40.51/0.9847
G	41.53/0.9882	39.63/0.9827	42.62/0.9896	43.32/0.9914	44.30/0.9924
B	38.61/0.9792	37.30/0.9705	39.57/0.9784	40.00/0.9830	40.65/0.9835
Color	39.08/0.9822	37.76/0.975	40.10/0.9824	40.61/0.9859	41.43/0.9869
Channel	IGD	NAT	CS	RI	MLRI
Channel	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM
R	39.66/0.9833	36.98/0.9709	39.04/0.9811	37.93/0.9762	38.87/0.9807
G	43.39/0.9913	39.43/0.9815	42.64/0.9902	40.99/0.9869	41.82/0.9889
B	40.02/0.9829	37.11/0.9692	39.16/0.9801	37.82/0.9734	38.85/0.9792
Color	40.63/0.9858	37.68/0.9738	39.91/0.9838	38.61/0.9788	39.57/0.9829
Channel	AICC	ARI	CNNCDM	Proposed-C	Proposed-A
Channel	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM
R	40.36/0.9857	39.16/0.9796	41.38/0.9862	42.51/0.9890	42.51/0.9891
G	43.09/0.9904	42.42/0.9884	44.84/0.9924	46.40/0.9943	46.41/0.9944
B	39.46/0.9797	38.98/0.9765	41.04/0.9841	42.23/0.9872	42.24/0.9873
Color	40.67/0.9852	39.87/0.9815	42.04/0.9876	43.25/0.9902	43.25/0.9903

Table 1. Objective evaluation of different algorithms on Kodak dataset

Channel	AP	AHD	DLMMSE	GBTF	LSSC
Channel	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM
R	32.79/0.9235	32.99/0.9267	34.02/0.9308	33.97/0.9373	36.02/0.9554
G	34.87/0.9417	36.97/0.9611	37.98/0.9631	37.34/0.9618	38.81/0.9733
B	31.98/0.8847	32.15/0.8819	33.03/0.8892	33.06/0.8977	34.71/0.9267
Color	33.01/0.9167	33.49/0.9232	34.46/0.9277	34.37/0.9323	36.15/0.9518
Channel	IGD	NAT	CS	RI	MLRI
Channel	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM
R	34.32/0.9409	36.27/0.9536	35.55/0.9564	36.1/0.9597	36.34/0.9605
G	37.37/0.9622	39.76/0.9729	38.84/0.9754	39.99/0.9797	39.90/0.9787
B	33.45/0.9045	34.39/0.9181	34.57/0.9301	35.37/0.9403	35.36/0.9388
Color	34.69/0.9359	36.20/0.9482	35.91/0.954	36.50/0.9599	36.62/0.9593
Channel	AICC	ARI	CNNCDM	Proposed-C	Proposed-A
Channel	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM
R	35.65/0.959	37.36/0.9666	39.14/0.9715	39.94/0.9753	40.19/0.9764
G	39.21/0.9756	40.67/0.9828	42.10/0.9844	42.72/0.9857	42.92/0.9861
B	34.33/0.9290	36.04/0.9433	37.30/0.9507	37.94/0.9550	38.06/0.9559
Color	35.85/0.9545	37.49/0.9642	38.97/0.9689	39.65/0.972	39.82/0.9728

Table 2. Objective evaluation of different algorithms on McMaster dataset

Jihui Yu, Xiaomin Yang. Double Branch Residual Network for Demosaicing[J]. Laser & Optoelectronics Progress, 2021, 58(2): 0210012

Download Citation