Classification of relative object size from parietooccipital hemodynamics using type-2 fuzzy sets

During the past two decade researchers have been exploring the mechanism of object shape and depth perception using EEG and fMRI. However, the underlying cortical process of perceiving different object sizes from a constant visual distance has never been explored. This paper provides a novel understanding of relative object size classification based on direct measure of parieto-occipital hemodynamics using functional near infrared spectroscopy (fNIRS). The cortical response is recorded from subjects engaged in visual perception task of relative object size. The signal is preprocessed (artifact removal) for construction of 176 features, which are thus reduced to 22 features using particle swarm optimization (PSO) technique. The reduced features are subsequently fed into an interval type -2 fuzzy set to classify the perceived objects (based on the underlying hemodynamic data) into three different classes: LARGE, MEDIUM and SMALL. Experimental analysis shows that the proposed feature-selection and classification framework attain higher classification accuracy which reaches over 87% in the classification of large objects. Analysis, further undertaken to know the underlying neurovascular mechanisms, reveals a distinct dorso-ventral shift (shall-medium-to-large) in parieto-occipital hemodynamic load which can be observed from the topographic brain activation. The average activation shifts are measured as 73.35 degrees in the right hemisphere compared to 93.71 degrees in the left hemisphere. The experimental outcomes could provide a novel measure in cortical hemodynamic features based perception of object size. In future, it could provide justification towards the visually challenged persons with perceptual difficulties.