Hierarchical Decision Tree (HDT) Approach for Image Annotation

Image annotation as a way to simplify searching images´ concepts, has been became an interesting research area in the recent years. In image annotation, the semantic concepts are added to images as some textual metadata. In this paper, we annotated images using decision trees which can select the most discriminatory features and are very interpretable. We made an ontology by organizing the semantic concepts hierarchically and then, by using a new decision tree construction algorithm which can deal with hierarchical structures, we reached to the more precise annotations. The main idea behind our approach is moving to the higher level and choosing more general concept, when inserting final nodes into the decision tree accompanies with no enough certainty. Simulation results confirmed that our approach illustrates more degree of accuracy in comparison with other decision tree construction algorithms, like ID3, which support only a linear relationship among concepts.