Automatic Annotation of Web Images combined with Learning of Contextual Knowledge

This paper introduces an approach for automatic image annotation which is based on contextual knowledge extracted from semantic rich web documents containing images. The knowledge itself is organized in ontologies and extended by learning algorithms for new contextual information. For this purpose, the contextual background of a picture is used for the annotation process and later in the image retrieval process. The paper shows the design of our system and how the different parts work together to enable and improve the annotation procedure. We created learning algorithms for harvesting new contextual information and thus improve context analysis. Finally, we evaluate our methods with a set of sports web pages.