Towards Automatic Image Annotation Supporting Document Understanding

The paper describes our research concerning image classification of types of graphics like plots, flow charts, illustrations and photos. Illustrations and photos are also classified into one of the following semantic classes: buildings, people, nature landscape, and interior. On this basis each image is annotated by its type and class. The key elements of the system - feature extraction and classification methods - are described in detail. A new classifier based on fuzzy logic was proposed. Moreover, we developed the Multi-Classifier, a hierarchical architecture encouraging the creation of hybrid classifiers tailored to the problem being solved. Experimental results of classification efficiency show that our approach is definitely worth further development.