基于内容的图像检索研究在过去的十几年里受到了广泛关注，取得了极大的发展。但是语义鸿沟问题的存在造成图像检索技术的应用效果距离用户的期望还有一定的差距。本项目将对图像检索中的语义特征表示及语义融合展开研究，以达到减小语义鸿沟的研究目的。本项目将对经典的词袋模型加以改进并提出一个语义图像检索架构。该架构利用视觉显著性模型及排序学习分别解决图像的语义特征表示及前景目标与背景区域的语义融合，以达到优化检索结果，满足用户高层语义检索需求的目标。本项目将提出基于色彩、纹理及形状等底层特征的语义特征表示方法及基于排序学习的语义融合方法。预计本项目的研究成果能广泛应用于网络搜索、智能交通、犯罪预防等领域。

Content-based image retrieval has been widely studied in the past decades and has already got great improvements. But the effect of the image retrieval has not met the requirements of the users due to the semantic gap. This project will focus on the semantic feature representation and semantic fusion to reduce the semantic gap. This project will modify the classical bag of words model and will propose a framework of semantic image retrieval. This framework will use saliency model for semantic feature representation and learning to rank model for semantic fusion to optimize the results of retrieval for meeting the requirements of semantic retrieval. This project will propose semantic feature representation method based on the low level features, such as color, texture and shape and the semantic fusion method based on learning to rank. The research results could be widely used in web search, intelligent transportation, crime prevention and so on.