Non Gaussian models for hyperspectral algorithm design and assessment
暂无分享,去创建一个
In this paper, we explore the use of elliptically contoured distributions (ECDs) to model the statistical variability of hyperspectral imaging (HSI) data. ECDs have the elliptical symmetry of the multivariate Gaussian distribution and therefore share most of its properties. However, the presence of additional parameters, allows to control the behavior of their tails to match the distribution of the data more accurately than the normal distribution. More specifically, the purpose of our paper is two fold. First, we provide a brief introduction to ECDs and their key properties. Second, we introduce the multivariate EC t-distribution and investigate its capability to accurately describe the joint statistics of HSI data from the HYDICE sensor.
[1] S. Kotz,et al. Symmetric Multivariate and Related Distributions , 1989 .
[2] Muralidhar Rangaswamy,et al. Computer generation of correlated non-Gaussian radar clutter , 1995 .
[3] A. Öztürk,et al. Non-Gaussian random vector identification using spherically invariant random processes , 1993 .