Automatic image annotation is an important and useful approach to narrow the semantic gap between visual features and semantics. However, it is time-consuming job since it extracting the visual features from a whole image to learn the relationship between low-level features and high-level semantic. In this paper, an image annotation method based on central region features reduction is proposed. Differ from the traditional annotation approach based on the whole image features, the proposed method analyze the central area which associate with the image semantics and only vision features of the area are extracted, then feature reduction based on Rough Set is used for getting the relationship between image visual features and semantics, lastly image annotation is executed. The experimental results show that the proposed method is effective and useful.