Single-pixel image reconstruction based on block compressive sensing and convolutional neural network

Single-pixel imaging (SPI) is an imaging technique that uses modulated light patterns and knowledge of the scene under view to obtain spatial information of the object. The combination of SPI and compressive sensing (CS) has enabled image reconstruction with fewer measurements. Typically, the reconstruction algorithm, such as basis pursuit, relies on the sparsity assumption in images. In this paper, we propose a SPI system based on block compressive sensing (BCS) and UNet-based convolutional neural network (CNN). Results show that our approach outperforms other competitive reconstruction algorithms. Moreover, by incorporating BCS, we can reconstruct images of any size above a certain smallest image size. In addition, we show that our model can reconstruct images obtained from an SPI setup while being priorly trained on natural images, which can be vastly different from the SPI images. This opens up opportunities for pretrained deep-learning models for BCS reconstruction of images from various domains.