Towards Optimal Binary Patterns for Compressive Terahertz Single-Pixel Imaging

Terahertz (THz) radiation's properties make it ideal for various imaging applications. However, creating simple, cost-effective, and high-resolution THz array detectors is challenging. Mechanical scanning is commonly used but creates a tradeoff between frame rate and resolution. Fortunately, Compressive Sensing (CS) offers a solution by reducing the required number of measurements needed compared to Shannon-Nyquist's sampling theory. CS-THz imaging is usually implemented using a single-pixel camera with spatial modulation patterns, mostly binary patterns. However, the non-uniform and diffraction propagation present in the THz range affects the mutual coherence of the resulting sensing matrices resulting in image reconstruction degradation. In this paper, we introduce an optimization procedure for generating binary patterns that consider THz diffraction and non-uniform illumination of the mask. The produced sensing matrices exhibit low coherence compared to other typical binary sensing matrices, resulting in a higher reconstruction performance than all others.