Real-Valued Network for Imaging through Multimode Fiber
- Resource Type
- Conference
- Authors
- Wang, Ziyu; Xiang, Yangyang; Lan, Mingying; Li, Junhui; Gao, Li; Yu, Song; Wu, Guohua; Jiang, Tianwei
- Source
- 2020 Conference on Lasers and Electro-Optics (CLEO) Lasers and Electro-Optics (CLEO), 2020 Conference on. :1-2 May, 2020
- Subject
- Photonics and Electrooptics
Training
Image reconstruction
Imaging
Artificial neural networks
Speckle
Optical fiber communication
Convolution
- Language
A real-valued neural network overwhelmingly outperforms its complex-valued counterpart in natural scene reconstruction through multimode fiber, suggesting that knowledge of intensity rather than phase information is enough, therefore brings much robustness and convenience in practice.