Multi-Focus Image Fusion Method Based on Cooperative Detection via a Deep Dense Convolutional Neural Network

Wei Yang; Liye Mei; Chuan Xu; Huan Zhang; Chuanwen Hu; Qianchuang Deng

doi:10.3788/LOP202259.2410004

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 24 >Page 2410004 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 24, 2410004 (2022)

Multi-Focus Image Fusion Method Based on Cooperative Detection via a Deep Dense Convolutional Neural Network

Wei Yang¹, Liye Mei², Chuan Xu³, Huan Zhang³, Chuanwen Hu^4、*, and Qianchuang Deng¹

Author Affiliations

¹School of Information Science and Engineering, Wuchang Shouyi University, Wuhan 430064, Hubei, China

²The Institute of Technological Sciences, Wuhan University, Wuhan 430072, Hubei, China

³School of Computer Science, Hubei University of Technology, Wuhan 430068, Hubei, China

⁴Zhejiang Academy of Surveying and Mapping, Hangzhou 311100, Zhejiang, China

show less

DOI: 10.3788/LOP202259.2410004 Cite this Article Set citation alerts

Wei Yang, Liye Mei, Chuan Xu, Huan Zhang, Chuanwen Hu, Qianchuang Deng. Multi-Focus Image Fusion Method Based on Cooperative Detection via a Deep Dense Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2410004 Copy Citation Text

show less

Overall framework of proposed fusion method. (a) Collaborative detection via densely connected convolutional neural networks; (b) multi-scale information extraction for pyramid pooling network

Fig. 1. Overall framework of proposed fusion method. (a) Collaborative detection via densely connected convolutional neural networks; (b) multi-scale information extraction for pyramid pooling network

Download full size

Fig. 2. Structure of deep dense block

Download full size

Fig. 3. Probabilistic decision diagram for multi-scale pyramid pooling collaborative detection. (a) Source image; (b) rough segmentation; (c) refined segmentation

Download full size

Fig. 4. Lytro-3 image fusion results. A is corresponding fusion results obtained by various methods, and B is difference images in pseudo color map form obtained by subtracting source image A from each fused image

Download full size

Fig. 5. Lytro-17 image fusion results. A is corresponding fusion results obtained by various methods, and B is difference images in pseudo color map form obtained by subtracting source image A from each fused image

Download full size

Fig. 6. More results of proposed method. The first to fourth and fifth to eighth columns show fusion results of 18 pairs of multi-focus images, respectively

Download full size

Image	Metrics	Fusion method
Image	Metrics	NSCT	GFF	IFM	CBF	DCHWT	CNN	PSPF	Proposed
Lytro-1	Q_MI	2.8165	3.6591	3.7414	3.3332	2.9543	3.7935	3.9901	4.0177
	Q_NCIE	0.8193	0.8301	0.8313	0.8255	0.8208	0.8318	0.8349	0.8354
	Q_AB/F	0.6004	0.7531	0.7405	0.7418	0.6945	0.7539	0.7560	0.7575
	Q_VIF	0.5432	0.7461	0.7037	0.6863	0.6476	0.7476	0.7452	0.7455
Lytro-3	Q_MI	2.8810	3.8132	4.0430	3.4832	3.1244	4.0501	4.1158	4.1351
	Q_NCIE	0.8203	0.8320	0.8355	0.8273	0.8229	0.8356	0.8368	0.8370
	Q_AB/F	0.5497	0.7437	0.7432	0.7311	0.6964	0.7466	0.7415	0.7455
	Q_VIF	0.4804	0.7546	0.7518	0.6936	0.6484	0.7593	0.7512	0.7571
Lytro-9	Q_MI	2.5304	3.8522	4.0807	3.6374	2.9118	4.0713	4.1162	4.1549
	Q_NCIE	0.8160	0.8319	0.8356	0.8288	0.8199	0.8355	0.8363	0.8367
	Q_AB/F	0.4625	0.8004	0.7992	0.7919	0.7448	0.8015	0.8010	0.8017
	Q_VIF	0.4193	0.7760	0.7703	0.7337	0.6546	0.7766	0.7793	0.7753
Lytro-17	Q_MI	2.4034	3.4225	3.6484	3.0606	2.7818	3.7062	3.8529	3.8604
	Q_NCIE	0.8143	0.8275	0.8301	0.8229	0.8191	0.8313	0.8328	0.8337
	Q_AB/F	0.5400	0.7660	0.7547	0.7529	0.7297	0.7695	0.7677	0.7693
	Q_VIF	0.4701	0.7123	0.6971	0.6496	0.6148	0.7227	0.7250	0.7270
Lytro-18	Q_MI	4.0753	4.8021	4.8594	4.5228	4.2013	4.9014	4.8296	4.9635
	Q_NCIE	0.8403	0.8525	0.8537	0.8476	0.8422	0.8545	0.8531	0.8557
	Q_AB/F	0.6282	0.7397	0.7352	0.7285	0.6877	0.7414	0.7332	0.7425
	Q_VIF	0.6226	0.8049	0.7916	0.7441	0.7232	0.8086	0.7891	0.7982

Table 1. Comparison of objective evaluation of fusion results

Metrics	Fusion method
Metrics	NSCT	GFF	IFM	CBF	DCHWT	CNN	PSPF	Proposed
Q_MI	3.1473	4.1211	4.2879	3.8211	3.3649	4.3210	4.4357	4.4656
Q_NCIE	0.8250	0.8395	0.8425	0.8344	0.8278	0.8430	0.8449	0.8459
Q_AB/F	0.5799	0.7601	0.7534	0.7528	0.7124	0.7618	0.7590	0.7669
Q_VIF	0.5132	0.7430	0.7233	0.6870	0.6465	0.7465	0.7406	0.7435

Table 2. Comparison of average objective evaluation of fusion results

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information