Analysis of Pulsed THz Imaging Using Optical Character Recognition

Using a reflection-based pulsed THz imaging system built upon our ErAs:GaAs photoconductive switch and a gated receiver, we quantify image quality at different detection bands (centered at 100, 400, and 600 GHz). Zero-bias Schottky diode detectors mounted in various waveguide sizes are used to tune the operational frequency bands of the imaging system, while the rest of the imaging system remains unchanged. The image quality is quantified by applying an optical character recognition (OCR) algorithm on THz images of 8-by-10 mm copper letters on a fiberglass substrate. Using the OCR success rate as a metric, we see a fivefold improvement in image quality from a 400 GHz to a 600 GHz imaging system, while our 100 GHz images do not produce any correct OCR results. In a comparison experiment performed at 600 GHz, the image signal-to-noise ratio (SNR) is degraded by placing increasing numbers of denim sheets (5.4 dB decrease in signal per layer) into the beam path. We find that the OCR success rate is roughly constant from one sheet to four sheets of denim (33-25 dB SNR) and then drops off sharply starting at five denim sheets.