Bayesian coupled-pixel terahertz-TDS imaging

This paper considers Terahertz (THz) sensing of contents on a single-layer object with time-domain spectroscopy (TDS) in a reflection mode. With an extremely small pulse width, the THz-TDS sensing is highly sensitive to non-ideal configurations such as a tilted or uneven sample, depth variation due to mechanical scanning operations, and intra-layer reflections. One consequence is sweep distortion resulting from slightly different time delay from different scanning points. In this paper, we follow an existing alternative minimization scheme and propose a simple modification to take into account, in addition to the binary nature of reflection coefficients, the two-dimensional block feature of the content (i.e., letters and shapes) in the sample. Preliminary results with synthetic datasets show the effectiveness of the proposed method.