Efficient Modeling of Spam Images

Image spam has become a real threat to email communication these days, since most prevalent content based spam filters can not efficiently detect them out, even when the latest OCR techniques are employed, spammers could compromise the system easily through text distortion and other obscuring skills. In this paper, we propose a novel and efficient image modeling approach for spam image classification, this content based statistical model does not rely on the availability of text information embedded in the image files, so that it is robust to obfuscations. Experimental results show that the proposed method can perform with good accuracy in practice.