METHOD OF THE CONSTRUCTION STATISTICAL MODELS OF OBJECTS BY MULTICHANNEL DATA

I.N. Kramskaya

The method of the construction multivariate statistical models of classification sign was considered, which are chosen as values of the components of color models (HSI, αβ). The control sample of the signs object classes were obtained by converting parts of an optical image corresponding to the objects of one class of data sets. Shape estimation of the distribution was based on statistical estimates of the coefficients of asymmetry and excess on the plane moments. To describe the components of the vector signs were selected normal, truncated normal distribution and a mixture of truncated normal probability density function. As a model of two-dimensional characters {H, S} and {α, β} used two-dimensional normal distribution, whose parameters were as statistical estimates of the expectation, standard deviation and correlation coefficients between the components. To refine the values of the parameters used by an optimization approach that provides the best fit of empirical and theoretical distributions.

Key words: color model, classification, multivariate statistical model, truncated normal distribution, parameters of the mixture.