2020
Common forms of eye-tracking analysis convey a subject's attention by grouping dense aggregations of gaze points into areas of interest. These areas of interest are often corroborated against pictorial figures such that the investigator can infer correlation of a subject's attention with a depicted object in a scene. However, while it is tempting to associate viewer interest with the images that underlay them, the identification of what constitutes legitimate pictorial elements can be contingent upon personal judgment, and so can differ drastically between subjects and the investigator. Accordingly, this research proposes an approach in which areas of interest are determined by data-driven methods using subject-specific gaze point information. The initial step of our method involved finding related clusters using a spatial cluster algorithm, and thereafter, each resultant group of gaze points was graphically outlined using a concave hull algorithm. We defined various typologies that describe the resolution of areas of interest both spatially and temporally.
data-driven AOIs
← back to research projects
PRECEDENT
Among the earliest attempts at quantifying areas of interest is found in Guy Thomas Buswell's How People Look at Pictures (1935). While visually unique, the division of an image into regular cells was, even by Buswell's admission, arbitrary.
©Original analysis over Leon Kroll's Morning on the Cape (c.1935)
PRECEDENT
A notable and early example of an investigator‐determined designation of areas of interest, from J.R. Antes' article "The time course of picture viewing" (1974), over Leon Kroll's Morning on the Cape (c. 1935). While the areas of each AOI encompass the same general size, some of their morphologies are significantly differently, and produced through a top-down and subjective determination.
Subject‐defined designation of Areas of Interest, using a clustering algorithm based on Ester, Kriegel, Sander, and Xu (1996), with perimeter articulated using a concave hull algorithm based loosely on Moreira and Santos (2007). Color and shading designating temporal sequence and neighborhood association, indicating clustering typologies.
