中文摘要：

提出一种自适应希尔伯特扫描方法用于解决图像检索中使用词袋模型丢失空间信息的问题。该方法通过分析特征在图像中的分布来计算在越来越精细的分辨率下每个希尔伯特路径的权重，从而为每张图像选择最优的扫描路径。探讨了基于希尔伯特扫描树的构建过程并对其优缺点进行了分析，该方法能够将图像特征的空间信息有效地加载到树的每个节点上。然后基于局部特征在图像空间的分布提出一种多层次的自适应希尔伯特扫描策略。得益于此方法，在之后为图像建立的基于希尔伯特树形结构上，物体的空间信息将被保存得更加准确，从而有利于快速重建物体轮廓。在Caltech-256数据库上进行了大量对比实验，实验结果表明该方法具有更高的检索准确率。

英文摘要：

One fundamental problem in large scale image retrieval with the bag-of-features is its lack of spatial information. An approach called adaptive Hilbert-scan depended on distribution of features in an image is proposed. This method computes weight of each Hilbert -scan at increasingly fine resolutions by analysis of feature distribution in the image,which is able to assign a suitable scanning path for each image. Hilbert-scan based tree structure is studied and its advantage ad disadvantage is analyzed. The method adds the spatial information of local features into each node of tree, furthermore a novel adaptive Hilbert-scan strategy with multi-level is designed, which is built on the distribution of features in image. Owing to merits of this method, spatial information of features will be preserved more precisely in Hilbert-scan based tree structures. Extensive experiments on Caltech-256 show the effectiveness of the method.