Why Do Humans Perform Perceptual Categorization

Humans perform perceptual categorization through the processes of visual recognition and attentional selection \cite{tva}. A perceptual categorization has the form ``$x$ belongs to $i$" (denoted by $ E(x,i) $, $ E(x,i) \in [0,1] $ ), where $x$ is an input element in the visual field to be categorized (called \textit{perceptual units}) and $ i $ is a \textit{perceptual category}. The collection of all perceptual units is denoted by $ S $ and the collection of all categories is denoted by $ R $.

In his work Bundesen discusses 3 types of perceptual categories, namely a color category, a shape category and a location category \cite{tva}. Our image retrieval model is founded mainly on color analysis. Therefore, we can follow the notation proposed in \cite{tva} with the assumption that perceptual categories are represented solely by colors.
\textit{Saliency} of $x$ to $i$ stands for the strength of sensory evidence that element $x$ belongs to a perceptual category $i$, denoted by $\eta(x,i)$. Every category $i$ is assigned a \textit{pertinence} value $\pi_{i}$, representing its current importance for a certain task $T$. Knowing pertinence and salience values we can derive an attentional weight of $x$