A new study, “VBSF: A Visual-Based Spam Filtering Technique for Obfuscated Emails”, offers a fresh solution. Conducted by Ali Hossary and Stefano Tomasin at the University of Padova, Italy, the research focuses on catching spam that hides in plain sight.
Researchers Ali Hossary and Stefano Tomasin from the University of Padova have developed a new way to catch spam that hides in plain sight. Traditional filters scan the text of an email; crafty spammers now evade those checks by burying keywords in faint colours or by swapping letters for look‑alike characters. The Padova team’s “visual‑based spam filter” (VBSF) treats each message the way a person does: it renders the email into an image so it can see the layout, colours and hidden text.
Once the email is rendered, VBSF runs two separate analyses. One pipeline uses optical character recognition to pull out the words and feed them to standard spam classifiers such as naïve Bayes and decision trees. The other pipeline looks at the image itself, feeding it to a convolutional neural network that learns to spot suspicious design choices. A meta‑classifier then combines both verdicts to decide if the message is spam or not. When tested on a hand‑curated mix of public and simulated spam datasets, this ensemble approach achieved more than 98 percent accuracy.
The idea matters because spammers are getting better at exploiting how emails render on screen. By mimicking human perception and analysing both the text and the visual presentation, VBSF closes a gap in traditional filters. While still a research prototype, it suggests that combining visual cues with textual analysis could help keep obfuscated spam out of inboxes.
The full paper is available on arXiv.
