ICA model of densities for images (ICADGGDD)

Abstract

In data analysis we use ICA is a basic tool, the aim is that find out a co-ordinate system where are the components of the data are independent.Mostly ICA method such as fastICA and Jointapproximation and diogonalization of eigen matrix (JADE), uses kurtosis as a metric of non gaussianity. But the assumption of kurtosis (fourth order cumulant) may not always satisfies practically. So there are one possible solution is to use skewness (third order cumulant) instead of kurtosis. In this paper we are going to introduce ICA based method, that approach is good for heavy-tailed (fourth order kurtosis) as well as asymmetric data (third order skewness).