What is it about?
Eye trackers record where a person is looking, but the recorded gaze points are often slightly wrong. This can happen because of calibration drift, small head movements, or other recording conditions. In reading studies, even a small error can make it unclear which word or line a person actually looked at. This work introduces GazeMorph, an AI-based method for correcting these errors after the data has been recorded. Instead of moving each gaze point separately, the method looks at both the pattern of recorded fixations and the structure of the page content. It then predicts smooth corrections that move the gaze points closer to the text or other visible content the person was likely viewing. The method was tested on reading data with different levels of simulated eye-tracking noise. Across all tested noise levels, GazeMorph reduced the average position error of fixations.
Featured Image
Photo by Josephine Vyeda on Unsplash
Why is it important?
Eye-tracking data is used to study reading, attention, user interfaces, accessibility, and human-computer interaction. However, if the recorded gaze points are inaccurate, researchers may draw the wrong conclusions about what people saw, read, or focused on. This work is important because it provides a way to improve gaze data after recording, complementing standard hardware calibration. It is especially useful in realistic settings where calibration can drift over time or where people move slightly during a session. By using the content on the screen as a guide, GazeMorph can make eye-tracking data more reliable without requiring a completely new recording setup. The results suggest that content-based correction can support more accurate reading studies and may contribute to future dynamic calibration systems that continuously improve eye-tracking quality during use.
Perspectives
As someone working with eye-tracking data, I see measurement error as one of the main barriers to using gaze information reliably outside highly controlled laboratory settings. This work was motivated by a practical question: can we use the content shown on the screen to better understand where a person was really looking? I hope this method helps researchers and developers make better use of noisy eye-tracking recordings, especially in reading research and human-computer interaction. In the longer term, I see this approach as a step toward eye-tracking systems that are more robust, easier to use, and less dependent on repeated manual calibration.
Almaz Shangareev
M. V. Lomonosov Moscow State University
Read the Original
This page is a summary of: GazeMorph: Reducing Eye-Tracking Measurement Error by Fixation-to-Content Alignment ETRA021, Proceedings of the ACM on Human-Computer Interaction, May 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3806035.
You can read the full text:
Contributors
The following have contributed to this page
