Many complex visual tasks can be decomposed into a sequence of simpler subtasks. Ullman (1984) suggested that the visual brain solves these subtasks by what he called “elemental operations”. More complex tasks can then be solved by routines, which are composed of a sequence of these elemental operations.
I will discuss connectedness detection, or “curve tracing” as a prototype of an elemental operation. I will review algorithms for the detection of connectedness from a computational, neurophysiological, and psychophysiological point of view. Our data indicate that recurrent connections in the visual cortex are used to solve this task. If a monkey has to delineate a connected image region, neurons in the visual cortex with a receptive field in this region enhance their firing rate, which spreads through the network until the whole connected region is labeled by an enhanced response. At a psychological level of description, this propagation of a firing rate enhancement corresponds to a shift of visual attention. Thus, visual attention spreads across image regions that connected to each other, until they all have been labeled by attention.
I will then discuss how the cortex solves more complex cognitive tasks that require the application of multiple elemental operations that together form a visual routine. It is possible to monitor the progression of such a visual routine by recording the activity of neurons in early areas of the visual cortex.