Saliency Detection of Light Field Image Based on Feature Fusion and Feedback Refinement

Xiao Liang; Huiping Deng; Sen Xiang; Jin Wu

doi:10.3788/LOP202259.2210006

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 22 >Page 2210006 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 22, 2210006 (2022)

Saliency Detection of Light Field Image Based on Feature Fusion and Feedback Refinement

Xiao Liang^1,2, Huiping Deng^1,2,*, Sen Xiang^1,2, and Jin Wu^1,2

Author Affiliations

¹School of Information Science and Engineering, Wuhan University of Science and Technology, Wuhan 430081, Hubei, China

²Engineering Research Center for Metallurgical Automation and Measurement Technology of Ministry of Education, Wuhan University of Science and Technology, Wuhan 430081, Hubei, China

show less

DOI: 10.3788/LOP202259.2210006 Cite this Article Set citation alerts

Xiao Liang, Huiping Deng, Sen Xiang, Jin Wu. Saliency Detection of Light Field Image Based on Feature Fusion and Feedback Refinement[J]. Laser & Optoelectronics Progress, 2022, 59(22): 2210006 Copy Citation Text

show less

Fig. 1. Overall architecture of proposed network

Download full size

Fig. 2. ECA network module

Download full size

Fig. 3. CFM network module^[13]

Download full size

Fig. 4. Comparison of PR curve results of differrent algorithms in (a) LFSD data set and (b) DUT-LF data set

Download full size

Fig. 5. Comparison of visual results of different algorithms in DUT-LF data set

Download full size

Fig. 6. Comparison of visual results of different algorithms in LFSD data set

Download full size

VGG-19					Dimensionality reduction
Layer	k×k-n	Input size	Output size	S	Layer	k×k-n	Input size	Output size	S
Conv1	3×3-64	256×256×3	256×256×64	1	Conv2	3×3-64	64×64×128	64×64×64	1
Maxpool	2×2	256×256×64	128×128×64	2	Conv3	3×3-64	32×32×256	32×32×64	1
Conv2	3×3-128	128×128×64	128×128×128	2	Conv4	3×3-64	16×16×512	16×16×64	1
Maxpool	2×2	128×128×128	64×64×128	2	Conv5	3×3-64	16×16×512	16×16×64	1
Conv3	3×3-256	64×64×128	64×64×256	1	Conv2		64×64×64	64×64×64
Maxpool	2×2	64×64×256	32×32×256	2	Conv3	Upsampling	32×32×64	64×64×64	1
Conv4	3×3-512	32×32×256	32×32×512	1	Conv4	Upsampling	16×16×64	64×64×64	1
Maxpool	2×2	32×32×512	16×16×512	2	Conv5	Upsampling	16×16×64	64×64×64	1
Conv5	3×3-512	16×16×256	16×16×512	1	Conv2-Conv5	Concat	64×64×64	64×64×（64×13）

Table 1. Network parameters of feature extraction module

DUT-LF data set					LFSD data set
Algorithm	F-measure	MAE	S-measure	E-measure	F-measure	MAE	S-measure	E-measure
LFS	0.533	0.227	0.585	0.742	0.735	0.205	0.681	0.773
RDFD	0.599	0.191	0.658	0.774	0.802	0.136	0.786	0.834
FPM	0.619	0.142	0.675	0.745	0.800	0.134	0.791	0.839
MWS	0.742	0.132	0.702	0.781	0.788	0.132	0.809	0.781
S2MA	0.753	0.102	0.787	0.816	0.803	0.094	0.837	0.863
DLSD	0.684	0.087	0.786	0.839	0.779	0.117	0.786	0.852
MAC	0.717	0.092	0.752	0.789	0.793	0.118	0.789	0.839
DLFS	0.868	0.070	0.852	0.905	0.715	0.147	0.737	0.806
MOLF	0.843	0.052	0.887	0.923	0.819	0.088	0.886	0.831
LF-Net	0.833	0.055	0.878	0.913	0.805	0.092	0.820	0.882
ER-Net	0.903	0.040	0.898	0.946	0.825	0.085	0.822	0.885
Proposed	0.871	0.049	0.890	0.913	0.812	0.088	0.889	0.843

Table 2. Comparison of index results of different algorithms in DUT-LF data set and LFSD data set

Algorithm	DUT-LF data set		LFSD data set
Algorithm	MAE	F-measure	MAE	F-measure
MOLF	0.059	0.801	0.088	0.786
ER-Net	0.048	0.870	0.089	0.805
Proposed	0.049	0.870	0.086	0.810

Table 3. Comparison of robustness results of different algorithms

Network structure	DUT-LF data set		LFSD data set
Network structure	MAE	F-measure	MAE	F-measure
Baseline	0.110	0.822	0.132	0.753
+SE and ConvLSTM	0.072	0.850	0.100	0.771
+ECA and ConvLSTM	0.061	0.863	0.094	0.798
+ECA and ConvLSTM	0.049	0.871	0.088	0.812
+Feedback refinement module	0.049	0.871	0.088	0.812

Table 4. Ablation experiments of different modules

Xiao Liang, Huiping Deng, Sen Xiang, Jin Wu. Saliency Detection of Light Field Image Based on Feature Fusion and Feedback Refinement[J]. Laser & Optoelectronics Progress, 2022, 59(22): 2210006

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享