Statistical Modeling for 3-D DFT Coefficients of Moving-Image Sequences and its Application to Denoising
Takashi Komatsu, Takahiro Saito · 2018
This paper presents simplification of the multidimensional 2-Component Spherically-Symmetric Gaussian Mixture (2-C S2GM) distribution model, developed for the statistical modeling of a moving- image sequence in the 3-DDFT domain; and this paper constructs a method to estimate model parameters of the simplified 2-C S2GM model from a noisy moving-image sequence to be denoised. To apply this statistical modeling to our previously proposed 3-D Mean-Separation-type Short-Time DFT (3-D MS2T-DFT) moving-image denoising method, this paper develops some key techniques such as an estimator of noise variance in the 3-D MS2T-DFT domain. Furthermore, through experimental denoising simulations, this paper demonstrates that our statistical modeling enhances utility and data-adaptability of our 3-D MS2T-DFT moving-image denoising method.