在Flickr图像共享网站上，大量无标签或者缺少标签的图像往往会因为标签信息的不完整，以致无法被有效地利用和检索。为了有效地进行图像检索，从Flickr用户经常会根据上传图像所隐含的主题而将其推荐到多个相关社群的特点出发，提出了一种新颖的基于社群隐含主题挖掘和多社群信息融合的自动图像标注算法。与传统的自动图像标注方法不同，该算法首先采用隐Dirichlet分配模型（latent Dirichlet allocation，LDA）对单个社群里的隐含主题（topic）进行挖掘，并利用隐含主题对由相似图像标签传播产生的初始“噪音”标签进行过滤；然后对同属于多个社群的图像，通过多社群信息融合来生成最终标注结果。实验结果显示了该新算法的有效性。

At photo sharing websites like Flickr, a lot of images can not be effectively used and retrieved due to lack of tags. In order to retrieve images effectively, this paper presents a novel social group latent topic mining and multi-group information fusion based automatic image annotation algorithm by exploiting the property that users in Flickr often recommend their uploaded pictures to associated social groups according to the hidden topics in each picture. Different from traditional automatic image annotation methods, this algorithm first adopts the latent Dirichlet allocation model to mine the latent topics in single social group and makes use of the hidden topics to filter initial noisy tags generated by tag propagation among similar images, then utilizes multi-group information fusion to generate the final annotations for images simultaneously belonging to multiple social groups. Experimental results show the effectiveness of this algorithm.