A demosaicing method for compressive color single-pixel imaging based on a generative adversarial network. (August 2022)

• Record Type: Journal Article
• Authors: Qu, Gang; Meng, Xiangfeng; Yin, Yongkai; Yang, Xiulun
• Is Part Of: Optics and lasers in engineering. Volume 155(2022)
• Publication Date: 2022-08
• Subjects: Color Single-pixel imaging -- Demosaicing -- Generative adversarial network; Lasers in engineering -- Periodicals; Optical measurements -- Periodicals; Optics -- Periodicals; Lasers en ingénierie -- Périodiques; Mesures optiques -- Périodiques; Optique -- Périodiques; 621.36605
• DOI: 10.1016/j.optlaseng.2022.107053
• Languages: English
• ISSNs: 0143-8166

Abstract

Highlights: A demosaicing method for compressive color single-pixel imaging based on a generative adversarial network is proposed. Bayer array is introduced into full color single-pixel imaging and the amount of data needs to be stored is reduced to one third. Then we utilize a generative adversarial network for demosaicing and achieve a better reconstruction result at lower sampling ratios compared with some state-of the-art demosaising mothods. Abstract: In conventional full color single-pixel imaging (SPI), multiple times measurement or multiple detectors are required to capture all channels' information of a color scene. The introduction of color filter array (CFA) provides a new idea for full color SPI and the sampling ratio is reduced to one third. However, the imaging quality and color fidelity largely depend on the demosaicing algorithm and when the sampling ratio decreases, the demosaicing result gets worse sharply or even totally fails. In this paper, we utilize a generative adversarial network (GAN) to implement color recovery specific for compressively sampled mosaic images reconstructed in SPI. Deep-learning based demosaicing method helps restore color SPI with higher imaging quality and color fidelity at lower sampling ratios compared with some state-of-the-art demosaicing algorithms. Both computer simulation and optical experiment have demonstrated the effectiveness and feasibility of the proposed method.