中文摘要：

为跨越高层语义概念与底层视觉特征之间的语义鸿沟，本研究提出一种新的图像自动标注方法。该方法首先采用灰度直方图方法对图像分割并提取图像区域的纹理特征，然后利用FCM算法中增大关联度高的特征权重更好地实现对分割后图像区域的聚类效果。最后改进贝叶斯分类器建立图像区域和语义概念间的关联模型，通过比较测试图像和训练图像间的最大相似度实现测试图像的自动标注。在Corel通用图像数据集上与其他几种方法进行了对比实验，实验结果表明改进后的标注方法优于传统标注方法。

英文摘要：

In order to cross the semantic gap between high-level semantic concepts and low-level visual features, a new method about automatic image annotation was proposed. First, the method of gray histogram was applied to segment images and texture features were extracted from image regions. Second, the greater weights could be distributed to rele- vant features compared with less relevant features for FCM algorithm, and better results of clustering could be achieved. Finally, the correlation model between keywords and clustering regions was established in accordance with the labeled images in the training sets by the approved Bayesian classification. The similarity between the testing images and the training images were calculated, and the maximal conditional probability to annotate the new image regions was achieved. The experiments on a standard Corel dataset compared with other methods showed that the approved labeling approach performed more accurately and effectively than traditional labeling methods.