A utilization of GMM for scientific images modeling
Jan Švihlík, Karel Fliegel · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2009
This paper deals with modeling of scientific and multimedia images in the wavelet domain. Images transformed into wavelet domain have a special shape of probability density function (PDF). Thus wavelet coefficients PDFs are usually modeled using generalized Laplacian PDF model (GLM), which is characterized by two parameters. The wavelet coefficients modeling can be more efficient, while the Gaussian mixture model (GMM) is utilized. GMM model is given by addition of at least two Gaussian PDFs with different standard deviations. There will be presented equation system derived by moment method for GMM models parameters estimation. The equation system was derived for an addition of two GMM models. So it is suitable for advanced denoising systems, where an addition of two GMM random variables is considered (e.g. dark current in astronomical images).