Ultrafast Imaging With Optical Encoding and Compressive Sensing

Abstract

Serial time-encoded amplified microscopy (STEAM) is an emerging technology, which enables an ultrafast phenomena to be captured at GHz scan rate. The tradeoff between high imaging speed and high spatial resolution remains a problem where the maximum scan rate is limited by the sampling rate of the digitizer and the temporal dispersion in the fiber to avoid data blending. In this paper, we address these limitations using state-of-the-art optimization algorithms under compressive sensing framework and establish the data acquisition model based on our proposed experimental setup by considering the effect of individual optical components such as laser spectral profile, encoding mask patterns, dispersion of the fiber, and optical noise in the system. We introduce two methods of alternating direction method of multipliers with total variation regularization (ADMM-TV) and discrete wavelet hard thresholding (DWT-Hrd) for STEAM-based imaging systems. Our results demonstrate that a 10-GHz scan rate can be achieved compared to the conventional 1-GHz microscopy imaging system while maintaining high image reconstruction quality in terms of structural similarity index measurement (SSIM). It is shown that among the two proposed optimization algorithms, ADMM-TV outperforms DWT-Hrd by 20% in SSIM measurements. Finally, it is shown that having 70%-80% light transmission through the mask reveals the optimum results in terms of reconstruction quality.