Abstract: Semantic auto annotation on Web image is an important method for huge amounts of images management and retrieval on Web,and the semantic mining from the images automatically and effectively is the key. The semantic lies not only in the image itself, but also and more importantly in its description texts. Further, the image semantic varies with the change of images or description knowledge. To address this issue, in this paper, based on domain ontology and different image descriptions,we propose an adaptive semantic annotation method for Web images. This method checks the impacts on image semantic from the description texts feature. It detects the semantic and extends keyword by domain ontology,and then based on a regression model to adaptively model the Web images' semantic distribution on its different description texts. A group of experiments are carried out on a real-world Web image data set and the experimental results show that our proposed method outperforms others is with excellent adaptivity.

Key words: Image annotation, Ontology, Semantic extension, Regression model