A System for Image Processing to Automatic Annotation

This paper presents a system for image automatic annotation considering the information extracted from the image through the segmentation and the boundary detection process. Using a domain ontology, we add semantic information to regions from the images which belong of to the domain. The technique proposed in the article uses a flexible hexagonal-grid structure which is over-lap on the image pixels and we extract for each segmented region the dominant color and prominent geometric properties. These characteristics are used in an inference algorithm to determine a decision tree. The instances of the decision tree together with the objects, corresponding to the low level features extracted from the images, are used for image annotation. The graph representation of images is the output of all the phases of system processing: the pre-processing phase, the segmentation step and finally, the automatic annotation. Our technique, which combines the visual feature descriptors with inference methods, has a good time complexity because it associates the semantic labels to regions by using the decision tree model.