Selection and Analysis of Objects in Multi-Threshold Image Processing

Selection and classification of objects according to certain objective criteria is one of the immense tasks for automated image analysis in computer vision systems. Here we revisit a recently suggested approach to the object selection problem based on choosing the optimal threshold to select a certain subclass of objects determined by objective criteria such as size, shape etc. While our earlier investigations mainly focused on the analysis of microscopic images, here we concentrate more on remote sensing applications. Besides revisiting the previously suggested method based on the size vs threshold histogram analysis, we also suggest several additional quantities that can be effectively used to find the best threshold once the empirical histograms exhibit several characteristic peaks. Moreover, similar approach could be useful for an automatic distinction between subclasses of objects with different morphology, thus resolving not only the selection but also the classification problem.