中文摘要：

视觉词包（Bag-of-visual-words, BoVW）模型是一种有效的图像分类方法。本文提出一种基于语义抽象的多层次决策（Multiple layer decision, MLD）方法,通过在BoVW 中引入抽象语义进行多层次扩展,采用语义保留方法生成具有语义的视觉词典,利用自底向上的方式逐层传递语义,训练上层语义分类器；分类时采用自顶向下方式逐层判断待测样本的类别。用标准数据集验证方法的分类性能。结果表明,本文提出的方法与主流分类方法相比具有更好的分类性能。

英文摘要：

Bag-of-visual-words （BoVW） is an effective method in image categorizing and retrieving task. A multiple layer decision method （MLD）, which introduces abstract semantics of image categories into BoVW to carry out middle-level and upper-level extensions, is proposed in this paper. Semantics is preserved at the stage of generating visual vocabulary, based on which classifiers are trained in a bottom-up way. Abstract semantics is transferred during the training step. After that, the category of a test image is estimated gradually by classifier through each layer in a top-down way. Experiments on standard datasets show that the proposed method achieves better performance compared with mainstream classification methods.