一个具有图像语义的物体分类系统的实现  

Realization of an Object Classifier on the Basis of Image Semantics

作　　者：孙季丰[1] 袁春林[1] 邱卫东[1] 余英林[1] 

机构地区：[1]华南理工大学电子与信息工程学院,广州510640

出　　处：《科学技术与工程》2008年第3期676-681,共6页Science Technology and Engineering

基　　金：国家自然科学基金(60372068)资助

摘　　要：通过提取图像的底层特征,将特征输入BP神经网,应用共轭梯度法对网络进行有监督训练,即将先验知识加入神经网络中;一旦训练成功,依靠网络出色的泛化能力,系统就能正确识别图像中的物体和各物体的位置信息,这样就实现了对图像语义的理解。实验验证结果表明该系统对特定测试图像集的理解正确率达到了100%。The research of image semantics is an active field in present time. It remains to be unsolved on the problem of ＂semantic gap＂ between the low-level image features and image semantics. A new method for the object classification is proposed, in which low-level features are extracted from some image, then inputed to a BP neural network and trained in the teacher＇ s guide. Once the training effectively finishs, by the excellent extending capacity of the net, the system can classify the objects in the image correctly, and give the location information of every object. In this way, the semantics of the image can get . Experimental results show the effectiveness of the proposed method on the object classification.

关 键 词：图像语义 底层特征 BP网络 位置信息 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]