基于Fisher权重改进的OB场景分类方法  

An improved OB scene classification method based on Fisher weight

作　　者：谢守志[1] 张磊[1] 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《应用科技》2014年第2期21-24,共4页Applied Science and Technology

摘　　要：目标库(object bank,OB)方法是一种使用了高水平语义特征的场景分类方法。针对OB方法使用过高维数的特征向量来表征图像的缺点,提出一种基于Fisher权重改进的OB方法。通过对没有进行空间金字塔处理的OB方法中的低维特征向量加以Fisher权重,使得新得到的特征向量具有极大化类间数据差异并且极小化类内数据差异的性质。实验表明,该方法与OB方法相比在分类效率上提高了10倍以上,并在分类的准确率上提高了3%左右。Object bank （OB） method is a scene classification method by using a high-level semantic features. A new improved OB scene classification method based on Fisher weight is proposed to overcome the disadvantage of using too high-dimensional feature vectors to express images in OB method. By adding the Fisher weight to the low-dimensional feature vector in OB with no spatial pyramid processing, the new feature vectors maximize the differences of the between-class data and minimize the differences of the within-class data. The experiments show that the improved method makes the classification efficiency 10 times higher than the OB method, and increases the classification accuracy by 3%.

关 键 词：OB方法 FISHER判别 场景分类 空间金字塔 支持向量机 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]