A Non-Content based Optimized Approach for Image Spam Detection

Image spam is a form of email spam that gained popularity in the last decade. Spammers have introduced the technique of image spam to bypass the traditional text-based spam filters. In this research, two novel techniques for image spam detection have been proposed and developed that are efficient as well have low computational complexity. The proposed techniques are non-content based. The first technique extracts low-level and high-level image properties and then feed these features into Random Forest classifier to perform classification. The second technique uses a Convolutional Neural Network where the raw image is directly fed into the CNN model to perform the classification between spam and non-spam images. This approach will remove the need for manual feature extraction thus reduce computational complexity. A comparative analysis is then performed between the two approaches.