Plug-and-play algorithm based on hybrid priors for single-pixel imaging

Abstract

Single-pixel imaging (SPI) is a powerful technique that captures scene data using a single-point detector with low cost, high signal-to-noise ratio and a broad spectral range. However, both the traditional model-driven methods relying on hand-crafted priors and the data-driven deep neural network methods still suffer from severe image degradation at ultralow sampling rates in practical applications. In this paper, we propose a novel hybrid-priors-driven plug-and-play (PnP) algorithm framework to significantly enhance the undersampling reconstruction performance. Unlike existing PnP methods that only rely on the deep denoising prior, but neglect the observation matrix physical prior information, we introduced an effective mask prior of the optical front-end to further boost the power of deep PnP framework. Extensive simulation and real data experimental results demonstrate that the proposed method achieves high quality image reconstruction at ultra-low sampling rates, and outperforms state-of-the-art PnP-based SPI algorithms, thereby facilitating the practical application of SPI technology.