Mixture Density Estimation In the Presence of Mixels

“Mixel” is a heterogeneous pixel which consists of more than two classification classes. This paper discusses the statistical properties of mixels, and also develops the theory of mixture density estimation in the presence of mixels. First we show the essential importance of the concept of stable distributions with regard to the analysis of mixels; next we derive a new statistical model called mixel distribution from a theoretical viewpoint. The application of the proposed mixture density estimation method to the classification of remote sensing images indicates the appropriateness of this method for the representation of long-tail distribution with using normal distributions and its mixel distributions only.