基于文本区域特征的图像型垃圾邮件过滤算法  

Image spam filtering based on features of text region

作　　者：耿技[1] 万明成[1] 程红蓉[1] 周俊怡[1] 

机构地区：[1]电子科技大学计算机科学与工程学院,成都610054

出　　处：《计算机应用》2008年第8期1904-1906,共3页journal of Computer Applications

基　　金：国家863计划项目(2006AA01Z41)

摘　　要：垃圾邮件图像中通常含有大量文本区域,且这些区域常含有较多区分能力强的特征。提出一种基于图像中文本区域特征的垃圾邮件图像识别算法。首先提取出图像中文本区域的特征,包括:文本区域数量和面积、色饱和度、文字数量和颜色数量,以及图像的一些属性特征如图像面积等;然后利用支持向量机分类算法来识别垃圾邮件图像。实验表明,对于真实的邮件图像集,算法能够识别出98.5%的垃圾邮件图像,且正确率超过98%。Text region embedded in most of spam images usually contains some discriminative features. An effective method based on the features of text regions to identify spain image was proposed. Firstly, the algorithm extracted the features of text regions that included the number and acreage, saturation, the number of characters and colors contained in text regions. Secondly, the spare images was classified by a support vector machine classifier. The experiment on real world data shows that the proposed algorithm can identify 98.5% of spam images and the precision is more than 98%.

关 键 词：图像型垃圾邮件 文本区域 垃圾邮件过滤 支持向量机 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]