Generative adversarial network‐based single‐pixel imaging. Issue 8 (10th May 2022)

• Record Type: Journal Article
• Authors: Zhao, Ming; Li, Fengqiang; Huo, Fengyue; Tian, Zhiming
• Is Part Of: Journal of the Society for Information Display. Volume 30:Issue 8(2022)
• Page Start: 648
• Page End: 656
• Publication Date: 2022-05-10
• Subjects: generative adversarial network -- image reconstruction -- learning‐based reconstruction -- single‐pixel imaging; Information display systems -- Periodicals; 621.38154205
• DOI: 10.1002/jsid.1128
• Languages: English
• ISSNs: 1071-0922

Abstract

Abstract: Single‐pixel imaging can reconstruct two‐dimensional images of a scene with only a single‐pixel detector. It has been widely used for imaging in nonvisible bandwidth (e.g., near‐infrared and X‐ray) where focal‐plane array sensors are challenging to be manufactured. In this paper, we propose a generative adversarial network‐based reconstruction algorithm for single‐pixel imaging, which demonstrates efficient reconstruction in high speed and higher reconstruction quality compared with other state‐of‐the‐art deep learning‐based reconstruction methods. We also co‐optimize sampling masks and reconstruction pipeline to enhance the performance. We verify the proposed method with both synthetic and real‐world experiments and demonstrate a good quality of reconstruction of a real‐world plaster using a 5% sampling rate. Abstract : A generative adversarial network‐based reconstruction algorithm is proposed to reconstruct the scene from single‐pixel measurements. The proposed network learned a set of new sampling mask and achieved good reconstruction with reduced measurements. Experimentally, we demonstrate a good quality of reconstruction using only 5% sampling rate.