In this paper, a model for multi-level image annotation that is performed in two phases is proposed. In the first phase, a Naïve Bayes classifier is used to classify low-level image features into elementary classes. In the second phase, a knowledge representation scheme based on Fuzzy Petri Net is used to expand the level of vocabulary and to include multi-level semantic concepts related to images into image annotations. In the paper, a knowledge representation scheme for outdoor image annotation is given. Procedures for determining concepts related to an image using fuzzy recognition and inheritance algorithms on a knowledge representation scheme are presented, as well as experimental results of image annotation.
