A study on visual attention modeling - A linear regression method based on EEG

In an increasingly knowledge based world, people are confronted with an explosion of information from the environment which must be viewed in restricted attention spans. Hence there is a need to investigate how best to model our Visual Attention (VA) with a view to allocate our attention efficiently. We use the color-word Stroop task combined with electroencephalogram (EEG) to model VA: subjects undertake the Stroop task and their EEG is recorded. This is in contrast to other studies that use techniques such as Event Related Potentials (ERP), Contextual Modeling Frameworks, eye movements and facial recognition. The paper presents a simple and useful model to recognize VA dynamically. We use the linear EEG features of different cortical fields as the main inference factors, and take the response time (RT) of the Stroop task as a metric to quantify subject performance. First, we obtain the most relevant EEG feature vectors from the recording, using a correlation analysis. Second, we use experimental data for training the VA model, using a regression method. Last, we then apply further experimental data to test the proposed model. The results from the tests conducted demonstrate that our model maps visual attention very closely.