Orientation specificity
The majority of cells found in area V4 exhibit tuning for simple attributes of the image, including spatial frequency or stimulus width, length and colour. Some cells in V4 do show selectivity to slightly more complex features (e.g. preference for a circular but not square shape, or the angle between 2 contours; Kobatake and Tanaka, 1994). The majority of V4 cells sensitive to simple or more complex features exhibit sensitivity to orientation (Desimone and Schein, 1987; Haenny and Schiller, 1988; Kobatake and Tanaka, 1994). Kobatake and Tanaka (1994) show an example of a V4 cell which is selective to the orientation of the optimal stimulus (a white bar presented at 315°) but generalises across size change [two sizes were tested, short (20%) and long (100%), see chapter VIE]. The main anatomical input into inferotemporal (IT) cortex is through V4 and it is therefore perhaps not surprising that cells in IT are also orientation sensitive (see Perrett and Oram, 1993; Oram and Penett, 1994a; Vogels and Orban, 1994).
For some features which do not have an axis of elongation such as a dark spot, a circular or square shape, and radially symmetrical patterns (e.g. high frequency Fourier Descriptors; Schwartz, Desimone et al., 1983; Gross, 1992) cells with preference for these features will automatically show tolerance of 90° or 180° rotation. In this sense, such feature detectors are orientation-invariant.
Tanaka et al. (1991) found that cells in the AIT which respond selectively to complex objects, such as faces and features, appear to be orientation selective. Ten AIT cells responsive to faces were tested for four orientations (0°, 90°, 180° and 270°). For these cells, the rotation of the optimal stimulus by 90° reduced the neuronal response by more than half. An additional 11 cells were reported which also exhibit orientation sensitivity, but were only tested qualitatively. The optimal orientation of the face stimulus varied from cell to cell, though the upright orientation was most commonly encountered (Tanaka et al., 1991).
Orientation generalisation
Physiological studies (Perrett et al., 1982, 1984, 1985; Hasselmo et al., 1989b) have shown that cells in the IT and STS cortex of macaque monkeys are selectively responsive to faces. In an initial study by Perrett et al. (1982) of selective cells in the STS, generalisation across object orientations appeared to be universal. All the face responsive cells studied (26) did not differentiate in response magnitude between upright, horizontal or inverted faces (Perrett et al., 1982). A closer examination of responses indicate that cell latencies can be affected by orientation even when response magnitude of the cells was similar for upright and inverted faces. Neural response latencies for 62% (16/26) of the cells were affected by the orientation of the image presented (Perrett et al., 1988). When the orientation of the image was changed from the upright orientation, the response latency of 10/26 cells were not affected. 15/26 cells responded with longer latencies to inverted or horizontally presented faces. Perrett at al. (1988) described this increased response latency as additional 'processing time' which may parallel increased RTs to identify faces or face configurations.
Hasselmo et al. (1989b) tested a small group of cells which responded to a particular movement (ventral flexion of the head; ‘head nodding’). These cells continued to respond to the head movements even when the image was rotated in the picture plane, i.e. upside-down. This process altered the retinal or viewer-centred co ordinate system, but not the object-centred co-ordinate system, since for the latter the position of the relative parts of the object are solely related to the object itself (i.e. the forehead moves towards the chest) and is therefore not dependent on the view. Their findings suggest that this small population of cells reflected the latter, the object- centred description of the movements. However, their results are limited in information about generalisation of rotation in the picture plane as they apply to stimulus motion.
In summary, physiological data shows that early visual areas (VI, V2, V4) exhibit orientation-specific coding of features of objects. Cells from these areas feed into more anterior visual processing areas (PIT, AIT), where cells are selectively responsive to progressively more complex features and object such as faces. From studies carried out so far, it appears that ail IT cells exhibit orientation specificity
Effect of Rotation 114 whether the ceils are selective for elementary features or more complex objects. Studies in the STS indicate the presence of cells which are selective for complex objects and which respond irrespective of the stimulus orientation. Interestingly, such cells exhibiting orientation-invariant responses to faces require extra time for the processing of unusual orientations of the stimuli.
It is pointed out, however, that for single cell studies described above only very few orientations were tested. Indeed, in the majority of instances only the neuronal responses to the upright and the inverted stimulus was investigated. The empirical study reported here was concerned with the extent to which cells in the STPa show orientation invariance (generalise across different orientations), or orientation specificity in their responses to heads, bodies or whole bodies presented in multiple orientations, often up to eight different orientations in the picture plane. This allows us to address the question of how orientation in the picture plane is processed by the visual system and whether this is done in a similar fashion as object view (see chapter VI and Perrett et al., 1991). In addition, neuronal response timing factors to different orientations of the stimulus for a cell population will be described and discussed.
Me t h o d
General single unit methods (see chapter IV) were applied to investigate the response pattern of cells in the STPa of the macaque monkey. All cells tested showed a significantly different response to the whole body than to both control stimuli and spontaneous activity (S/A). For each cell the optimal perspective view was first defined for the upright body (see chapter VI), and subsequently the cell was presented with that view of the whole body in different orientations. Note that the entire body stimulus was used for testing purposes. This is because cells selectively responsive to only one component part usually (see chapter V) responded to the entire body significantly better than to S/A and control objects. In addition, it is assumed that neuronal response selectivity to a particular orientation is held constant across different body parts (and the entire body) as seen with view. That is, if e.g. a ‘head alone’ cell is selectively responsive to a particular orientation of the head presented in isolation, then it is expected to find that the entire body presented in the same orientation as the head alone will elicit a cell response of the same magnitude (or larger, see chapter V) than
the response to the head presented in isolation. This makes sense, if one follows the argument from chapter V, that component parts are associated with each other. It is unlikely, therefore, to find e.g. a figure where the head is upright in respect to gravity and the body component is inverted.