Feature Analysis in Inferior Temporal Cortex

As noted earlier, the pattern of receptor firing rates on the retina constitutes the initial representation of an object. In a like manner, the center-surround ganglion cells and simple and complex cells in V1 can be considered increasingly complex object representations, with each cell coding for a particular feature (e.g., an edge tilted at a particular orientation) of the object. How are objects represented in later visual cortical areas?

Axons from neurons in V1 project to several other cortical areas, including (logically enough) areas V2 and V3. From here, visual processing separates into two streams. One, commonly referred to as the "where" pathway, flows dorsally, ending in the poster ior parietal lobe, and determines the spatial locations and/or functional aspects of visual stimuli. The second stream, which flows ventrally toward the inferior temporal lobe, is responsible for determining object identity. The present discussion is confined to this latter "what" pathway, because we have defined object perception as the process of recognizing the identity of an object.

Relatively little is known about the function of areas V2 and V3, in part because, whereas V1 is easily exposed by surgically removing portions of the skulls of laboratory animals, V2 and V3 are hidden in folds of the occipital cortex. It is possible that these areas, as well as area V4, to which neurons from V2 and V3 project, provide the medium for further stages of representation in the ventral processing stream. We know more about the functional characteristics of cells in the inferior temporal lobe (area IT), which receives input from V3 and V4 and is the last purely visual processing area in the ventral stream. Researchers recording from single cells in various parts of IT have found neurons that respond best to complicated patterns of light. For example, one study found a neuron that responded with a rapid firing rate to a plastic model of a pineapple (Fig. 5A). Further investigation revealed that the neuron responded equally well to the leaves of the pineapple isolated from the rest of the object (Fig. 5B), and the researchers eventually established that the neuron's most rapid firing rate occurred in response to an eight-pointed star (Fig. 5C). This type of neuron was dubbed the "elaborate cell," suggesting a more complicated building block for representations of whole objects, itself constructed from sets of complex cells in earlier visual areas.

Other studies have found neurons in IT that respond best to whole objects, most notably faces. Some neurons are remarkably selective, consistently responding at their fastest rate only when a particular familiar face is presented (e.g., the face of one of the experimenters). These findings seem, on the surface, to support the concept of "grandmother cells," single neurons that are each responsible for signalling the presence of a single object, for example, one's grandmother. However, several considerations indicate that this notion is almost certainly misguided. Whereas a particular neuron might respond best to a particular face, it will also respond fairly well to other faces. Furthermore, most face-selective cells in IT (as well as most elaborate cells) respond best when the image of the face is of a particular pose (e.g., profile or full frontal), is in a particular orientation (most cells

Figures Schematic results from single-cell recording studies of neurons in cortical area IT. The vertical lines below each image show action potentials generated by the neuron while the image is shown to the animal subject (this figure illustrates the results of a study using macaque monkeys as subjects). Faster firing rates presumably indicate a greater preference by the cell for a given image. Results shown in (A) indicate that the cell responds better to an image of a pineapple than to other fruits. (B) indicates that the cell appears to be responding to the leafy part of the pineapple, not to the shape or texture of the bottom portion. Finally, (C) shows that the cell responds best of all to an eight-sided star shape.

Figures Schematic results from single-cell recording studies of neurons in cortical area IT. The vertical lines below each image show action potentials generated by the neuron while the image is shown to the animal subject (this figure illustrates the results of a study using macaque monkeys as subjects). Faster firing rates presumably indicate a greater preference by the cell for a given image. Results shown in (A) indicate that the cell responds better to an image of a pineapple than to other fruits. (B) indicates that the cell appears to be responding to the leafy part of the pineapple, not to the shape or texture of the bottom portion. Finally, (C) shows that the cell responds best of all to an eight-sided star shape.

respond best to upright faces, although some respond better to misoriented views), and takes up a particular retinal size (corresponding to a certain viewing distance). Thus, a neuron that responds better to grandmother's than auntie's face when they are presented from the front, tilted 30°, and viewed from 2m, might reverse its preference if the faces were shown in profile, upright, and viewed from 10 m. The pattern of responses over a large number of IT neurons, rather than single cells, must be considered for object identity to be determined.

Because the majority of IT cells that have been found to respond best to a particular object also respond best when the object is seen from a particular orientation, view, and distance, the single-cell recording literature would seem, on the surface, to support view-based over structural description theories. That is, if area IT is the site of high-level object representations and neurons in IT show viewpoint-dependent responses, one might conclude that high-level representations are view-based. Again, though, patterns of responses over a population of neurons, rather than single neurons alone, represent objects, and it is possible that an appropriately selected population of IT neurons might respond to objects in a viewpoint-invariant manner. Furthermore, some single neurons do respond to objects across a wide range of views, and the response properties of other neurons are completely unknown (indeed, in most studies only a relatively small proportion of neurons are found to selectively respond to any one object). At present, we can conclude that single-cell recording studies are more consistent with view-based than with structural description theories, but it is entirely possible that future studies may reveal evidence for structural descriptions (or some sort of currently unimagined hybrid representation) in area IT.

Before we leave the topic of single-cell recording, another feature of the receptive fields of IT cells should be noted: the receptive field size (the area of the retina to which a neuron responds) is greatly expanded for these cells compared to simple and complex cells in V1. V1 neurons, in turn, have larger receptive fields than retinal ganglion cells; more generally, there is a trend for receptive field sizes to increase as information gets farther along in the ventral processing stream. A natural consequence of this trend is the effective solution to translation constancy. Whereas a particular simple cell "looking" for a 45° oriented edge will only respond if SOGI's head falls within a small area, an IT cell that responds best to the SOGI as a whole will fire regardless of where in the visual field the object is located. However, it is important to note that, whereas the large receptive fields of IT cells indicate that translation constancy has been achieved by this point in processing, we do not yet know exactly how these cells come to respond to objects in any part of the visual field.

Was this article helpful?

0 0
Conquering Fear In The 21th Century

Conquering Fear In The 21th Century

The Ultimate Guide To Overcoming Fear And Getting Breakthroughs. Fear is without doubt among the strongest and most influential emotional responses we have, and it may act as both a protective and destructive force depending upon the situation.

Get My Free Ebook


Post a comment