A behavioral approach to supervised classification

Related to the papers by Rasson et al. (1993) and Orban-Ferauge (1993), the present author summarizes a new behavioral answer to the problem of classification of pixels from their spectral coordinates. The graphical illustration of the mathematical foundations of the method is shown, attempting to make clear what is the basic notion of the intensity estimate on each point (which should not be confused with the notion of density). The author also discusses the comparison between the present new approach and the classical ones by the presentation of color printed classifications and confusion matrices. The author synthesizes the main conclusions resulting from the comparison of several methods of classification (including interactive maximum likelihood, expert systems and neural networks) and suggests a few ideas for discussion and future research.< >