Open World Logo Detection and Classification

Malicious webpages will often display brand logos to feign legitimacy; however, these brand logos are often distorted from official versions. The goal of this project is to build a classifier that recognizes and identifies logos in a screenshot of a legitimate or malicious webpage. The classifier must also be able to quickly begin classifying new and modified logos as they begin to be used for both legitimate and malicious purposes (e.g., “open-set” learning). This classifier may be used within a suite of tools to assess the legitimacy of a webpage, as well as to help track brand misuse.