Fundus Image Segmentation Using Decision Trees

In this work, the study on feature selection effectiveness using decision trees was conducted. Our previous work allowed us to identify the most informative texture characteristics using the selection traits technology. The technique made it possible to carry out intelligent analysis of features using color subspaces when solving the problem of selecting regions of interest. In this paper, decision trees were constructed using texture features that were used for the previously proposed technology. Decision trees selected new informative signs, providing higher accuracy. The most effective number of features is 3 when the window size is more than 15. For windows of dimension 12, decision trees using more than 6 texture features provided accuracy above 98.