Serial versus parallel theories of perception

The research on visual cortical neurons was at first thought to support serial hierarchical theories of perception (Selfridge, 1959), in which perception is thought to proceed in a sequence of stages, starting at the retina and ending (presumably) somewhere in the cortex, with information flowing in just one direction. Such frameworks can be called ‘hierarchical’ because a unit in each successive stage takes input from several units in the preceding stage. This kind of organization could be likened to the Catholic church, in which several parish priests report to a bishop, several bishops to a cardinal, and several cardinals to the pope. In the same way, general features of the retinal image, such as lines, were thought to be extracted by early visual processing, while whole complex objects were recognized later in the sequence by the analysis of combinations of these features. For example, the capital letter ‘A’ contains a horizontal line and two opposite diagonals, the letter ‘E’ contains three horizontals and a vertical, and so on. These letters can therefore be defined with respect to a combination of their elementary perceptual features. Representations of corners, squares, and then three-dimensional cubes, were thought to be built up by combining the outputs of these early feature detectors to form more complex object detectors in ‘higher’ regions, such as the cortex of the inferior temporal lobe. However, more recently there has been an increasing emphasis on the parallel organization of the cortex (Livingstone & Hubel, 1987). So in V1, M and P cell signals (projected from the magno and parvo components of the retina, respectively) arrive in different layers of the cortex. These messages are processed in V1 and are then carried by axons out of V1 and into several adjacent regions of the cortex, called V2, V3 and V5. In V2, Livingstone and Hubel argued that the M and P signals are kept separate in different columns of cells. Consistent with our previous discussion these columns represent information about motion and distance (magno system) and colour (parvo system), respectively. This theory became complicated by Livingstone and Hubel’s description of activity in a third type of column in V2, where the cells receive converging input from the magno and parvo systems. They suggested that these columns are used for spatial pattern analysis. However there are problems with this scheme. For example, Livingstone and Hubel claimed that images in which the different regions are red and green, but all of the same brightness appear flat. They attributed this to the insensitivity of cells in the magno/depth system to differences purely in hue, which are detected primarily by the parvo system. Quantitative studies, however, found that perceived depth is not reduced at all in such images (Troscianko et al., 1991). It appears, then, that depth percepts can be derived from both magno and parvo information, though not necessarily equally well at all distances (Tyler, 1990). In fact, there are many more visual areas in the cerebral cortex than are shown in figure 8.12. Some two dozen or so have now been discovered by neuroanatomists and by brain imaging studies .The functions of these areas are still being studied intensively by physiologists and psychologists, and we do not yet have the complete picture. Zeki (1993) has put forward the most influential theory of cortical visual functioning. According to this scheme, area V3 is important for analysing stimulus shape from luminance or motion cues, V4 is important for the perception of colour and for recognising shape from colour information, and V5 is critical for the perception of coherent motion. But this theory is still controversial. Recent physiological studies have found fewer differences between the properties of the various cortical areas, emphasizing that many areas co-operate in the performance of any given task. For example, Lennie (1998) points out that most information flow in the brain is from V1 to V2 to V4, and that area V4 is not sp cialized for colour in particular, but for finding edges and shapes from any cue or feature. Lennie argues that only the small stream through V5 is specialized, to monitor image motion generated by self-movement of the body and eyes (optic flow). This would therefore be the area activated in the illusion of selfmotion we experience when the other train moves, as described at the very beginning of this chapter.