Semantic Annotations of Image Based on Mutual Information and Constrained Clustering
基于互信息约束聚类的图像语义标注

An image annotation method based on mutual information and constrained clustering is proposed. We utilized the semantic constraint to improve information bottleneck method,which employed to cluster the segmented region.Then relationships between image semantic concept and clustering regions are established. Toward the un-annotated image, a new method is proposed to calculate the conditional probability of each semantic concept,while considering the prior knowledge of training images and low-level features of the segmented regions.Finally,the image region semantics are automatically annotated by keywords with maximal conditional probability. The proposed method has been implemented and tested on an image database with about 500 images. The experimental results show that the effectiveness of the proposed method outperforms other approaches.