Detecting visual feature importance via tree classifiers. An experience

n this work we present an experience of using tree classifiers as a tool of multivariate exploratory analysis. It refers to a problem of image analysis, and specifically to that of identifying low-level visual features suitable to capture salient aspects of the semantic content of the images. The problem is addressed by studying the discrimination capability of the features towards classes of images semantically defined. The classes considered are the photograph, graphic and text classes.