如何面对人们日常生活中接触到的，尤其是互联网上数量激增的图像进行有

效的分类，已经成为研究的新热点。虽然现有的图像分类技术已经取得不错的性

能，但是它们还存在着一些问题。一是大部分现有的图像分类算法都是基于图像

的底层特征，’无法解决图像分类中的“语义鸿沟”问题；二是，大多数图像分类算

法总是忽视图像中部分与部分之间的空间关系。-How to face contact with people' s daily lives, especially surge in the number of images on the Internet for effective classification, has become the new hot spot. Although the existing image classification techniques have made a good performance, but they still exist some problems. First, most of the existing image classification algorithms are based on the underlying characteristics of the image, ' can not solve image classification in the " semantic gap"　problem　Second, most of the image classification algorithm is always ignored and some parts of the image between spatial relationships.