Markov chain Monte Carlo sampling based terahertz holography image denoising.

Terahertz digital holography has attracted much attention in recent years. This technology combines the strong transmittance of terahertz and the unique features of digital holography. Nonetheless, the low clearness of the images captured has hampered the popularization of this imaging technique. In this paper, we perform a digital image denoising technique on our multiframe superposed images. The noise suppression model is concluded as Bayesian least squares estimation and is solved with Markov chain Monte Carlo (MCMC) sampling. In this algorithm, a weighted mean filter with a Gaussian kernel is first applied to the noisy image, and then by nonlinear contrast transform, the contrast of the image is restored to the former level. By randomly walking on the preprocessed image, the MCMC-based filter keeps collecting samples, assigning them weights by similarity assessment, and constructs multiple sample sequences. Finally, these sequences are used to estimate the value of each pixel. Our algorithm shares some good qualities with nonlocal means filtering and the algorithm based on conditional sampling proposed by Wong et al. [Opt. Express18, 8338 (2010)10.1364/OE.18.008338OPEXFF1094-4087], such as good uniformity, and, moreover, reveals better performance in structure preservation, as shown in numerical comparison using the structural similarity index measurement and the peak signal-to-noise ratio.