单尺度词袋模型图像分类方法  

Single-scale image classification employing Bag-of-Words model

作　　者：陈凯[1] 肖国强[1] 潘珍[1] 李正浩[2] 

机构地区：[1]西南大学计算机与信息科学学院,重庆400715 [2]重庆大学光电技术及系统教育部重点实验室,重庆400044

出　　处：《计算机应用研究》2011年第10期3986-3988,共3页Application Research of Computers

基　　金：国家自然科学基金资助项目(61003203);重庆市自然科学基金资助项目(CSTC2010BB2230)

摘　　要：针对基于SIFT特征描述的图像分类方法需构造多尺度极值空间,运算耗时且部分极值点无直观视觉意义,提出一种新型的图像分类方法。该方法通过网格直接提取单尺度SIFT特征,并对局部特征进行单尺度词袋模型描述。由于单尺度SIFT无须构造多尺度空间且保留了更多的全局信息,从而极大地降低了计算复杂度且使分类正确率得到显著提升。实验结果表明,提出的单尺度SIFT比常规SIFT所形成的词袋模型在分类正确率上有明显提高。The general image classification methods relying on SIFT feature description need to construct multi-scale space,thus it is not only time-consuming but also irrelevant to visual sense.This paper proposed a new image classification method.It directly extracted single-scale SIFT features based grid,and described the features employing Bag-of-Words（BOW） model afterwards.Because single-scale SIFT need not build multi-scale space and retains more global information,the proposed method could reduce the computational complexity substantially and improved the classification accuracy significantly.Experimental results illustrate that compared with the standard SIFT based BOW model,the classification accuracy of BOW model formed from single-scale SIFT is significantly improved.

关 键 词：图像分类 单尺度SIFT 视觉单词 词袋模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]