Extracting relevant visual properties of made-man objects with monadic and pair-wise evaluation: Mixture modeling and parzen kernels infer attention

This work is a study of visual perception of object's design. While participant are assessing object's design visual properties, eye movements are recorded, in the aim to extract attentionnal grabbers, and identify knowledge-guidance and perceptual-guidance on information uptake. Monadic and paire-wise presentation are used and analyzed with a variety of data treatments applied on eye movements records. Exhaustive panel of methods for eye movements analyze is build and classified in two main categories: spatial density distribution and evolutional time impacted areas of interest. Mixture Gaussian algorithm is applied for density clustering. Effects of task, scene, and subjects properties are analyzed with these tools. It appears that for this type of cluttered scene, and for evaluation tasks, the eyes seem to be guided by object 's representation more than by task relevant features.