A single-pixel imaging method via low-resolution illumination patterns

• Published in: Optics communications Vol. 565; p. 130665
• Main Authors: He, Ziqiang, Dai, Shaosheng, Liu, Jinsong
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 15.08.2024
• Subjects: Compressed sampling; Deep learning; Image reconstruction; Single-pixel imaging; Deep learning; Single-pixel imaging; Compressed sampling; Image reconstruction
• ISSN: 0030-4018; 1873-0310
• DOI: 10.1016/j.optcom.2024.130665

Focusing on the field of single-pixel imaging for reconstructing high-resolution images, most existing methods end-to-end employ illumination patterns and reconstructed images of the same resolution, resulting in a singular imaging resolution. Crucially, in actual test environments, enhancing the resolution of illuminating patterns and increasing sampling times tends to introduce more disturbance information, consequently affecting the clarity of the reconstructed images. Therefore, this study proposes a single-pixel imaging method via low-resolution illuminating patterns. The method employs a small number of low-resolution illuminating patterns for compressed sampling of the targe, acquiring rough light signals with less disturbance interference, and directly generating a high-resolution image, termed Coarse Sampling Single-Pixel Imaging (CSSPI). The results show that the CSSPI method proposed in this paper demonstrates a 40.00% reduction in the amount of sampling information when compared to OMP, TVAL3, DLBOGI, and DSPINet. In simulation experiments with a salt-and-pepper noise ratio of 0.00–15.00%, or actual tests with smoke concentration ranging from 0 to 32 g/cm3, the CSSPI employing the low-resolution illumination pattern scheme for 4-times resolution reconstruction, consistently demonstrates relatively better imaging stability and clarity. The introduction of this method could potentially provide a modest contribution to the study of stable imaging for single-pixel technology in practical applications. •Low-resolution illumination pattern has better anti-interference capabilities.•Reconstructing image resolution is not constrained by illumination pattern.•Fewer sampling times help control introduction of disturbance information and shorten sampling time.