Application of wavelet domain Markov random field model in THz image processing

The main challenges for free-space terahertz (THz) imaging are known to be atmospheric loss, moisture absorption, low radiation power; and concequently, low signal-to-noise ratio (SNR). The need to have higher power radiation sources; faster data acquisition times remain major obstacles for high image quality. In this paper, the current state of research and applications was analyzed, as well as the future development of THz imaging technology was predicted. The basic principle of synthetic aperture radar (SAR) imaging and THz compressed sensing (CS) imaging was expounded. The THz image features of the two imaging methods were analyzed. The denoising effects of THz simulation images among the Wiener2, ddencmp, Donoho and the wavelet coefficients Amplitude Asymptotically Optimal (AAO) algorithm were also compared, qualitatively. A Markov random field(MRF) model for THz image denoising was presented, in order to capture the characteristics of scale space, with better scale wavelet coefficients in the wavelet domain. The image's MRF model was established and the energy functions which were used for image denoising and the two states of each wavelet coefficient were introduced, in non-stationary regions: one state corresponded to the image features such as edge, while another state was related to the stationary region image. The Expectation Maximization (EM) algorithm was used to estimate the parameters of the mixture model, along with the Bayes Preliminary rule to determine the ideal image wavelet coefficients contraction factor. The denoising algorithm of the Hidden Markov Models in Wavelet Domain (HMMWD) was tested, with excellent simulation results that show the WDHMM to be more effective.