A homograph phishing attack uses character script similarities to create and register fake domains of well-known brands to trick users into visiting these websites. This paper proposes a new visual inspection-based approach, Phish-EYE, for homograph phishing domain detection through binary visualization of domain names using TensorFlow. Phish-EYE can determine if a website domain is homograph or legitimate. Phish-EYE classifies homographs and non-homograph domains by converting the URL string into a binary string and using the byte class colour scheme to create an image. The TensorFlow machine learning framework is used to classify homographs and legitimate domains because it can learn image differences well. The Phish-EYE experiments have shown reliable and promising results. The average performance of Phish-EYE in identifying the homograph domain results in a 96.51% sensitivity (true positive rate) and 98.01% accuracy. The Phish-EYE performance demonstrates the feasibility and usefulness of leveraging the domain binary visuals to identify homograph phishing scams.