基于决策融合的图像自动标注方法

Method of automatic image annotation based on decision fusion

结合多媒体描述接口(MPEG-7)和MM(Mixture Model)混合模型,实现了基于决策融合的图像自动标注。在图像标注过程中,分别利用颜色描述子和纹理描述子为每个主题下的图像建立MM混合模型,实现低层视觉特征到高层语义空间的映射,利用局部决策融合方式融合在颜色和纹理MM混合模型下的标注结果,实现图像自动标注。通过在corel图像数据集上的实验,表明提出的局部决策融合方式能更充分利用图像的颜色和纹理信息,提高了图像标注性能。

A method for automatic image annotation based on decision fusion is proposed combining the Multimedia Descrip- tion Interface （MPEG-7） and MM （Mixture Model）. In the process of image annotation, two independent MM mixture models are estimated for the images belonging to a theme and mapping is setted up from low-level features to high-level semantics space. Automatic image annotation is achieved by fusing the annotation results from color and text MM mixture model in the way of local decision fusion. The way of local decision fusion is proven to utilize fully the color feature and texture feature and improve the performance of image annotation by the experiments on the image data sets.