Scanpath Comparison for Visual Search Analysis
Eye Tracking | Classification | Art
Challenge
Users are given a visual task, while I study their scanpaths. Scanpaths are sequences of eye movements. After recording the scanpaths with an eye tracker, I want to analyze the recorded data; and my research question was: It is possible to classify users into different groups only by using scanpaths?
Solution
I took a concrete example: Given scanpath from art experts and laymen looking at famous paintings, I tried to classify them according to their expertise. After temporal binning, feature discretization, binary representation, data reshaping, and dimensionality reduction, I could classify viewers with an accuracy of 64.64%. Interestingly, I also found that a user’s scanpath is characteristic for that user, i.e., the user can be recognized among others, so a user seems to have a particular fingerprint.
My Role
This was my master thesis ; it was a cooperation of HPI with the Computer Engineering Department of University of Tuebingen. I had the opportunity to apply techniques from Machine Learning and Data Mining, techniques that had not been used for this before.
Technology
MATLAB, Graphical Models, Hidden Markov Models, SVD, Data Cleansing, R
