Introduction of the bootstrap resampling in the generalized mixture estimation

Abstract

In this paper, we propose to introduce the bootstrap resampling technique in the generalized mixture estimation. The generalized aspect comes from the use of the probability density function (pdf) estimation coming from the Pearson system. The bootstrap sample is constructed by randomly selecting a small representative set of pixels from the original image. The application of the Bootstrapped Generalized Mixture Expectation Maximization algorithm BGMEM led us to define a new empirical criterion of representativity of the sample. We give some simulation results for the determination of the empirical criterion. We validate our criterion by the application of the algorithm to the problem of unsupervised image classification.