基于可控金字塔子带能量特征的文种识别方法  

LANGUAGE IDENTIFICATION METHOD BASED ON SUB-BAND ENERGY FEATURES OF STEERABLE PYRAMID

作　　者：顾立娟[1] 邵命山[1] 郝玉保[2] 

机构地区：[1]武汉军械士官学校 [2]75719部队

出　　处：《计算机应用与软件》2011年第3期91-94,共4页Computer Applications and Software

基　　金：国防建设基金项目

摘　　要：为了更精确地描述文本图像的纹理特征进行文种识别,提出了一种利用可控金字塔对图像进行分解并提取子带系数的特征量进行文种识别的方法。通过在两个不同质量图像库上进行的对比实验,证实了基于子带能量均值和标准差的方法具有更高的识别准确率,而基于广义高斯模型的方法对文本行倾斜更具有鲁棒性。In order to describe the texture features of text images in script more accurately for language identification,we proposed a language identification method by utilizing steerable pyramid to decompose the image and extract the eigenvalue of sub-band coefficients.Through a series of comparative experiments conducted on two image databases with different quality,we have confirmed that the method based on mean and standard deviation features of sub-band energy has higher recognition accuracy while the method based on Generalized Gaussian Model（GGM） is more robust to the slant of text lines.

关 键 词：文种识别 可控金字塔变换 广义高斯模型 纹理特征 文本图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]