Detecting Image Spam Using Image Texture Features

Abstract

Filtering image email spam is considered to be a challenging problem because spammers keep modifying the images being used in their campaigns by employing different obfuscation techniques. Therefore, preventing text recognition using Optical Character Recognition (OCR) tools and imposing additional challenges in filtering such type of spam. In this paper, we propose an image spam filtering technique, called Image Texture Analysis-Based Image Spam Filtering (ITA-ISF), that makes use of low-level image features for image characterization. We evaluate the performance of several machine learning-based classifiers and compare their performance in filtering image spam based on low-level image texture features. These classifiers are: C4.5 Decision Tree (DT), Support Vector Machine (SVM), Multilayer Perception (MP), Naïve Bays (NB), Bayesian Network (BN), and Random Forest (RF). Our experimental studies based on two publicly available datasets show that the RF classifier outperforms all other classifiers with an average precision, recall, accuracy, and F-measure of 98.6%.