Binarization for Single-Pixel Imaging Using Error Diffusion with the Maximum and Minimum Values

Single-pixel imaging (SPI) differs from conventional imaging techniques because it captures images using a single-pixel sensor, thereby enabling imaging in low-light environments across a wide range of wavelengths. However, one of the major challenges in SPI is that obtaining high-quality reconstructed images requires a large number of mask patterns, which substantially increases both the computational cost and projection time. In addition, high-speed projection requires a spatial light modulator capable of displaying binary patterns, such as a digital micromirror device. This study proposes a simple and efficient binarization method for SPI. The proposed method uses an error diffusion method with the maximum and minimum values of the original mask patterns. To demonstrate the effectiveness of the proposed method, it is applied to deep-learning-based SPI, in which convolutional kernels are employed as mask patterns. The proposed method was applied to the trained mask patterns and did not require binarization in the training stage, demonstrating that it can achieve high-quality reconstructed images.