基于改进二值模式的图书文档图像分类  

Document Image Classification Based on Improved Local Binary Patterns

作　　者：张敏[1] 

机构地区：[1]河南理工大学测绘与国土信息工程学院,河南焦作454003

出　　处：《红外技术》2014年第10期827-831,共5页Infrared Technology

基　　金：河南省国际合作项目;编号:134300510057

摘　　要：在分析针对局部二值模式进行降维方法的基础上,提出了一种改进的局部二值模式描述符,并用于图书文档图像分类。新方法首先将原局部邻域划分为多个4-正交邻域,然后统计4-正交邻域二值化后所包含的"1"的个数作为特征,最后融合所有4-正交邻域特征进行图像分类。采用广泛应用的纹理图像库、前视红外目标图像库和图书文档图像库进行实验,结果表明,新方法的特征维数不但明显降低,而且还取得了较高的分类准确率。Based on the analysis of the methods to reduce the dimensions of the local binary pattern（LBP）, a new operator called the orthogonal combination of local binary number（denoted as OC-LBN） is proposed for document image classification. Firstly, the local neighborhood is divided into different 4-orthogonal-neighbors, and the binary number of ＂1＂ in each 4-orthogonal-neighbor is used as its feature. Then, the features of all the 4-orthogonal-neighbor are combined together as region description. Experimental results obtained from texture, forward-Looking infrared and document image databases demonstrate that the proposed method can get the best performance of the methods mentioned in the paper.

关 键 词：图像分类 局部二值模式 纹理分析 降维 

分 类 号：TN219[电子电信—物理电子学]