Abstract:
This research is aimed at the development of a versatile tool that
couples gaze tracking technology with image matching techniques in an
efficient manner. The developed software can easily identify and highlight
areas on a computer screen that a user is focusing on. The tool effectively
handles the zooming feature through optimized feature matching and template
matching techniques. It has diverse applications, and its utility is tested via
experiments on histopathology slides and PDF documents. Annotating large
medical datasets manually is a lethargic and time-consuming process but such
properly annotated datasets are vital for creating accurate machine learning
models. This dissertation addresses this challenge by streamlining the
annotation process. Experiments performed on histopathology images reveal
that the software correctly identifies areas of the medical slides that are of
interest with a high accuracy. It highlights these regions through a heatmap so
that they can be easily annotated or tagged. Similarly, experiments performed
on PDF documents showcase that the tool can easily detect the text that is
being read by the user. Overall, feature matching outperforms template
matching in both kinds of experiments by achieving an accuracy, precision
and F1-score of over 0.9. Beyond medical imaging and assistance in data annotation, the tool has potential for a wide range of applications. It can be
used for detecting bias in the review of academic papers. Moreover, it can
provide insights into user behavior and preferences that can help in the
formulation of effective marketing strategies. Thus, the tool can have
numerous applications as it seamlessly blends gaze tracking technology with
image matching algorithms.
