Image Super-Resolution Reconstruction Method Based on Self-Attention Deep Network

Zihan Chen; Haobo Wu; Haodong Pei; Rong Chen; Jiaxin Hu; Hengtong Shi

doi:10.3788/LOP202158.0410013

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 4 >Page 0410013 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 4, 0410013 (2021)

Image Super-Resolution Reconstruction Method Based on Self-Attention Deep Network

Zihan Chen¹, Haobo Wu^2、*, Haodong Pei^3、*, Rong Chen¹, Jiaxin Hu¹, and Hengtong Shi¹

Author Affiliations

¹Futian Power Supply Bureau, Shenzhen Power Supply Bureau Co., Ltd., Shenzhen 518001, China

²School of Electronic Engineering, Xidian University, Xi'an, Shaanxi 710071, China

³Key Laboratory of Intelligent Infrared Perception, Shanghai Institute of Technical Physics, Chinese Academy of Sciences, Shanghai 200083, China

show less

DOI: 10.3788/LOP202158.0410013 Cite this Article Set citation alerts

Zihan Chen, Haobo Wu, Haodong Pei, Rong Chen, Jiaxin Hu, Hengtong Shi. Image Super-Resolution Reconstruction Method Based on Self-Attention Deep Network[J]. Laser & Optoelectronics Progress, 2021, 58(4): 0410013 Copy Citation Text

show less

Fig. 1. Overall structure of SADeepNet

Download full size

Fig. 2. Structure of self-attention layer

Download full size

Fig. 3. Plane images reconstructed by different methods. (a) Low-resolution input image with enlarged display; (b) Bicubic; (c) SRCNN; (d) FSRCNN; (e) SADeepNet; (f) high-resolution original image

Download full size

Fig. 4. Chair images reconstructed by different methods. (a) Low-resolution input image with enlarged display; (b) Bicubic; (c) SRCNN; (d) FSRCNN; (e) SADeepNet; (f) high-resolution original image

Download full size

Fig. 5. Butterfly images reconstructed by different methods. (a) Low-resolution input image with enlarged display; (b) Bicubic; (c) SRCNN; (d) FSRCNN; (e) SADeepNet; (f) high-resolution original image

Download full size

Dataset	Bicubic	SRCNN	FSRCNN	SADeepNet
Plane(×2)	19.99	21.16	21.37	22.10
Plane(×3)	18.39	19.21	20.08	21.22
Chair(×2)	18.40	18.55	18.96	19.67
Chair(×3)	17.32	17.92	18.04	18.46
Butterfly(×2)	16.38	16.81	17.13	17.91
Butterfly(×3)	16.23	16.72	17.07	17.75

Table 1. PSNR comparison of different methods unit: dB

Dataset	Bicubic	SRCNN	FSRCNN	SADeepNet
Plane(×2)	0.66	0.70	0.71	0.73
Plane(×3)	0.62	0.64	0.66	0.69
Chair(×2)	0.55	0.57	0.59	0.63
Chair(×3)	0.52	0.52	0.55	0.58
Butterfly(×2)	0.56	0.57	0.59	0.64
Butterfly(×3)	0.53	0.55	0.58	0.60

Table 2. SSIM comparison of different methods

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information