In order to obtain a general solution, we considered it
important that the input images cover wide ranges of sizes,
colours and number of objects. In order to make our features
comparable, we use normalisation: all extracted features are
represented in the form of ratios and percentages.