Support Vector Machines for Image Spam Analysis

Abstract

Email is one of the most common forms of digital communication. Spam is unsolicited bulk email, while image spam consists of spam text embedded inside an image. Image spam is used as a means to evade textbased spam filters, and hence image spam poses a threat to email-based communication. In this research, we analyze image spam detection using support vector machines (SVMs), which we train on a wide variety of image features. We use a linear SVM to quantify the relative importance of the features under consideration. We also develop and analyze a realistic "challenge" dataset that illustrates the limitations of current image spam detection techniques. 2 BACKGROUND In this section, we briefly consider spam detection techniques. Then we discuss research that is most clo

Deep convolutional neural network based image spam classification

“…In [15], the performance of PCA and SVM based image spam classifiers are studied on two datasets in which the Linear and Radial Basis Function SVM models achieved better results. The first dataset is developed by [13] and the second improved dataset is developed by [16] which cannot be detected properly by PCA and SVM based approach.…”

“…This dataset is developed as a challenge dataset in work [16] in order to test the performance of image spam models with more advanced spam images. It contains a total of 1,029 spam images that are generated by embedding spam text in ham images as shown in Fig.…”

Classifying Classic Ciphers using Machine Learning

“…Combining two or machine learning models is discussed by a previous research, where malware files are classified by considering them as images, thus converting it to a image classification problem using CNNs [16]. Another previous research to detect metamorphic malware combines scores from different techniques, namely, HMMs, Simple Substitution Distance (SSD) and Opcode Graph Similarity (OGS) and uses a SVM to classify these scores [17].…”

Deep Learning for Image Spam Detection

“…This dataset was created by Chavda et al [4] using image processing techniques on spam images to make them appear more like a ham image. A public corpus named Spam Archieve [18] consists of only spam images.…”

“…Previous research into image spam detection has shown that some types of image spam can be detected with high accuracy. For example, in [4,5] These experiments serve two purpose. First, we can determine an effective strategy in the ''cold start'' case, that is, in the case where the training data is severely limited.…”