The original idea was to perform a "sliding-window" correlation on the scanpath data in order to measure similarity across subjects. We've settled on converting each scanpath to a dimension-reduced 3D brick and correlating those. It's a little more computationally intensive, but allows for some imprecision across subjects in all three dimensions.
The original idea was to perform a "sliding-window" correlation on the scanpath data in order to measure similarity across subjects. We've settled on converting each scanpath to a dimension-reduced 3D brick and correlating those. It's a little more computationally intensive, but allows for some imprecision across subjects in all three dimensions.