Noise variance estimation of three-dimensional mean separation type ST-DFT coefficients.

Abstract

Three-dimensional mean value separation type short time DFT is effective for movie restoration. We modeled the ST-DFT coefficient as two component spherically-symmetric Gaussian mixture distribution. We proposed the estimation approach of the model parameters using the moment method. If we know the amount of noise mixing in the movie, parameter estimation procedure can executed without iteration. In this purpose, we propose the estimation method of noise variance value in moving-image by using three dimension DFT coefficients.