Abstract: In recent years, in the bag-of-words model, the position relationship between visual words is almost completely abandoned, existing redundant information can’t express this kind of image effectively. Based on the problems which the relative position relations between traditional visual words are under-utilization and semantic information of visual words is not clear, this paper proposes visual phrases based on support region detection for image representation. Some support regions can be obtained, which play an important role in image classification by support region detection. Then visual words from support regions are formed into visual phrases which are used for image classification. Finally the contrast experiment is done in the UIUC-Sports8 and Scene-15 databases, the experimental results show that the proposed method has a good image classification performance.

Key words: bag-of-words, support region detection, visual phrases, image representation

摘要： 近年来，基于bag-of-words模型的图像表示方法由于丢弃了视觉词汇之间的空间位置关系，且存在冗余信息，从而不能有效地表示该类图像。针对传统词袋模型视觉词汇之间相对位置关系利用不足，以及语义信息不明确的问题，提出采用基于支持区域的视觉短语来表示图像。通过支持区域探测得到图像中对分类起重要作用的支持区域，然后对支持区域上的视觉词进行空间建模得到视觉短语用于分类。最后在标准数据集UIUC-Sports8图像库和Scene-15图像库上进行对比实验，实验结果表明该算法具有良好的图像分类性能。

关键词: 词袋, 支持区域探测, 视觉短语, 图像表示