In MR Jenkin and LR Harris (eds.), Vision and Attention (pp. 169-188). New York: Springer. 2001. [pdf]
The negative definition of mid-level vision contained in this description reflected a rather large gap in our understanding of visual processing: How could the here-and-now descriptions of the low levels combine with the atemporal knowledge of the high levels to produce our perception of the surrounding world?
A number of experimental and theoretical efforts have been made over the past few decades to solve this "mid-level crisis". One of the more recent of these is based on the phenomenon of change blindness--the difficulty in seeing a large change in a scene when the transients accompanying that change no longer convey information about its location (Rensink, O'Regan, & Clark, 1997; Rensink, 2000a). Phenomenologically, this effect is quite striking: the change typically is not seen for several seconds, after which it suddenly snaps into awareness. During the time the change remains "invisible", there is an apparent disconnection of the low-level descriptions (which respond to the change) from subjective visual experience (which does not). As such, this effect would seem to have the potential to help us understand how mid-level mechanisms might knit low- and high-level processes into a coherent representation of our surroundings.
It is argued here that this potential can indeed be realized, and that change blindness can teach us much about the nature of mid-level vision. A number of studies are first reviewed showing that the perception of a scene does not involve a steady buildup of detailed representation: rather, it is a dynamic process, with focused attention playing one of the main roles, viz., forming coherent object representations whenever needed. It is then argued that change blindness can also shed considerable light on the nature of focused attention itself, such as its speed, capacity, selectivity, and ability to bind together visual properties into coherent structures.