Abstract: A novel method of image annotation and retrieval based on improved k-means algorithm is proposed.First，image segmentation is applied to the training image；an improved k-means algorithm is proposed to cluster the segmented region.The improved k-means algorithm first uses the genetic clustering algorithm to determine the number of cluster k，and then chooses the clustering centers.At the time of image annotation，the correlation between image semantic concept and region cluster is established，and then it is used as a priori knowledge and is combined with the lower feature of the region to annotate the un-annotated image.The system has been tested on an image database of about 1 000 images.The experiment results show that the proposed approach has a good retrieval performance.