New estimators for tail heaviness parameter of generalized Gaussian distribution

The estimation task of tail heaviness parameter p of generalized Gaussian distribution is considered. Two simple p estimators on basis of kurtosis and its robust modification, percentile coefficient of kurtosis, are proposed and their accuracy is analyzed. Based on the obtained results, two new p estimators are developed employing hard threshold and combination schemes. As a practical application of the proposed estimators the task of studying statistical properties of test image DCT-coefficients is considered.