DOG_I: An Annotation System for Images of Dog Breeds

摘要

DOG_I (Dog OntoloGy Image annotator) is a complete and fully automated semantic annotation system for images of dog breeds. Annotation relies on feature extraction and on associating low-level features with image concepts in an ontology. Because general purpose ontologies for all image types are not yet available, we choose the problem of annotating images of dog breeds as a case study and for the evaluation of our methodology. Nevertheless, DOG_I can be adapted to more image types provided that an ontology for a new image domain becomes available. Therefore, DOG_I offers an ideal test-bed for experimentation and sets the grounds for the annotation and evaluation of virtually any image type. Evaluation results are realized using images collected from the Web. Almost 95% of the test images is correctly annotated (i.e., DOG_I identified their class correctly). DOG_I is accessible on the Internet.