中文摘要：

提出一种新的框架用于改进传统词袋模型效率较低的问题。该方法建立在通过小波变换获取的低尺度图像表示上,利用在低尺度图像上提取单尺度的SIFT特征,建立低尺度视觉词典。由于大幅度减少了图像初始特征维数,该方法可以快速建立视觉词典,并且有效地降低后续图像分类所花费的时间。通过对Caltech101数据集全部8 677张图像的分类测试显示,该方法可以在保证分类性能的同时,有效地提升基于传统词袋模型的图像分类效率。实验结果表明,该方法可以全面提升金字塔匹配的词袋模型分类性能和分类效率,普遍用于传统词袋模型及其衍生方法。

英文摘要：

This paper proposes a new framework to improve the efficiency of visual bag-of-words model in large scale image classification. The method is based on the low scale image representation obtained by wavelet transform, and the low scale visual dictionary is built by extracting the SIFT features on the low scale image. Since the feature dimension is reduced, the method can quickly generate the visual dictionary and minimize the time of image classification process. The results of comparison experiments on the 8 677 images of Caltech 101 show that the proposed method can effectively improve the classification performance and efficiency of the traditional visual bag-of-words model and the Pyramid-BOW model.