In search of relevant things: a novel approach for image analysis

As our eyes search for meaning in what we see, our brains switch between two distinct methods of observation processing. Bottom-up processing collects sensory inputs as raw signals to the brain (e.g., colours, textures, and geometry). In contrast, top-down processing recognises patterns to define meaningful interpretations from those signals. Since Noton and Stark’s 1970s research into the significance of visual pattern processing, contemporary studies are finding that the seemingly volatile jerk-like movements of the eye, known as saccades, are not random but are governed by shared biological visual effects. 

Further investigation into how similarly we view a shared stimulus, and the biological conditions that would cause us to see something differently, can potentially uncover new avenues for understanding cognition. For example, vertigo has been measured by examining involuntary eye movements, known as nystagmus. Dementia can also cause the eye to scan stimuli differently, revealing eye gaze patterns for viewers experiencing cognitive decline. The dynamic between the visuo-spatial sketchpad, which stores visual information for manipulation, can assist in measuring cognitive load, which manages how the human brain stores that information. Alternatively, research has shown that eye gaze patterns can be markers for expertise during an experiment evaluating medical images. 

This research aims to both develop computational methods to measure the perception of a shared stimulus and demonstrate how that research can uncover cognitive function biomarkers. Gaze patterns are challenging to measure due to their spatial complexity when using topological dimensions in Euclidean geometry. However, using fractal geometry, they display self-similarity at different scales within a defined window of ranges, making them well suited to measurement using quantitators such as the Higuchi Fractal Dimension. The top three significant improvements fractal geometry can provide include the improvement of eye gaze representation behaviours in machine learning feature spaces, improvement of eye gaze comparison and measurement methods, and improvement of behavioural difference stratification between bottom-up and top-down processes within distinct cohorts. This research will impact clinical applications such as disease screening for conditions like dementia and autism and teaching applications such as radiology and crime Ascene investigation.