Abstract
Despite the great popularity and convenience of eye tracking experiments, a well-known problem is its limited power in reading the internal mental status. For example, the raw fixation data do not tell whether the user is aware of the contents where fixations landed. Most studies simply assumed that what is "fixated" is what is "seen", which is not always true. To predict visual awareness, here we zoomed in each fixation in a visual search task and built a model to predict user awareness in a fixation-by-fixation basis. We first conducted a visual search task where 20 visual search arrays containing separate objects were designed and eye movement data from 11 subjects were collected. The visual search paradigm provided the ground truth of visual awareness (based on whether subjects pressed the key to indicate target detection) for learning a prediction model. We then extracted a number of eye fixation features including previous and current fixation duration, normalized pupil size, fixation spatial density, microsaccade number, and previous saccade magnitude. These features were fed into a linear Support Vector Machine (SVM) to predict user awareness of target detection. After cross validation, we obtained accuracy levels of 92.2% for non-recognized fixations on target and 86.5% for recognized fixations on target. SVM also identifies the most important features in the prediction as fixation spatial density, current fixation duration, and previous saccade magnitude. The results show that the proposed eye fixation features and the machine learning technique can predict visual awareness in a high confidence level. The effective prediction of visual awareness suggests that eye fixations contain rich information and is useful in revealing cognitive status.
Meeting abstract presented at VSS 2014