SPI-CGAN: Single-pixel imaging based on conditional generative adversarial network

Abstract

Imaging quality at extremely sampling ratios is a key research topic in the field of single-pixel imaging (SPI). Although conventional methods can be used to reconstruct the object images at low sampling ratios, the reconstructed image is still visually unclear. To solve this problem, an SPI model based on a conditional generative adversarial network (SPI-CGAN) is proposed to achieve an end-to-end reconstruction of the object images and improve the image quality of the reconstruction at extremely low sampling ratios. To improve the stability of the training model, the objective function is composed of multiple loss functions. The validity of the model is verified through simulation data, which do not require a long time to collect. The optimized SPI-CGAN can reconstruct a sharp image edge at extremely low sampling ratios. Compared with a linear iterative method and a nonlinear iterative method, the proposed method performs better in terms of the quantitative indicators used.