BASES EPISTEMOLÓGICAS

The overall response pattern of the entire cell population which was selectively responsive to the static head/body was investigated. From the PSTH it is evident that cells respond equally fast to the whole body in its preferred view than to the whole body in its non-preferred view. If mental rotation (Shepard and Metzler, 1971) occurred at a neuronal level, i.e. a rotation process was applied to the incoming image or the representation stored, then one would expect to find that the response onset to the best view would be earlier than the response onset to the worst view (the view to be transformed/rotated). This time difference in latency could be argued to be due to the mental rotation process. However, no such time difference was observed and therefore, the data presented here does not support models suggesting mental rotation.

Effect of Rotation 94

Chapter VII

Ef f e c t o f r o t a t io n o n o b je c t r e c o g n it io n

In t r o d u c t io n

Rotating an object in the picture plane does not influence the relative positions of the object's parts visible from one perspective view. The overall direction of the parts, however, relative to the observer, and relative to the gravitational and scene- based axes does change. For example, if the whole body is viewed upright and from the left profile, then the nose points to the left of the viewer. If, however, this view of the body is inverted, even though the body remains in left profile view, the nose now points towards the right of the viewer. If a representation of an object present in higher visual areas can be activated independent of the view and orientation relative to the observer, then such a representation will be referred to as object-centred. On the other hand, if the representation of an object is preferentially activated by a specific orientation, then the representation will be referred to as viewer-centred.

Behavioural studies in Humans

Short term memory

Simultaneous matching of shapes

Shepard and Metzler (1971) presented subjects with two shapes simultaneously. Subjects were asked to determine whether the shapes were the same or were mirror images of each other. All subjects reported that the method used, was to imagine a three-dimensional rotation of one of the objects and thereby matching it to second image. It was found that reaction times (RTs) to perfomi such a task increased as the angle between the first and second image increased. This increase of reaction time was taken as an estimate of the time to ‘mentally rotate’ one image in such a way that it could be directly matched against the other image. Hence, the larger the angle of rotation between the two images, the longer the reaction time needed for mental rotation to be completed.

Shepard and Metzler (1971) therefore suggested that mental rotation occurred at a speed of 60° per second. It should be noted, however, that inverted images of most

familiar objects do not require 3 seconds for recognition. Thus, mental rotation phenomena (at this speed) appear too slow for general object recognition. Moreover, Corballis and Nagourney (1978) argued that if mental rotations did occur, one would have to expect that if the orientation of the viewed object is changed relative to the orientation of the stored representation, there would be a linear function relating stimulus orientation to recognition times. RT/orientation functions are, however, often non-monotonic, with 180° rotations (inverted images) being processed faster than 120° rotations (Jolicoeur, 1985; McMullen and Jolicoeur, 1990; 1992).

Successive matching of shapes

Tarr and Pinker (1989) asked subjects to learn a series of letter-like characters. The subjects were then shown one character presented in a different orientation from the trained character and had to recognise the image as the same or as a mirror reflection of the trained image. Initially, the greater the angle of rotation between the learned image and the test image, the longer the RTs, however, a practice effect did arise (see below). They also found that recognition of characters that were a mirror reflection from the trained images, was performed with equal efficiency independent of the orientation of the stimulus. That is, the time taken to match an upright letter-like object to its upright mirror image was the same as the time taken to match an inverted letter-like object to its inverted mirror image. It should be noted that mirror images preserve the orientation of many of the object’s component features, thus matching mirror images and matching across a rotation of >45° may involve fundamentally different processes.

Long term memory

Identification and naming tasks

Jolicoeur (1985) measured recognition times of line and colour drawings of common objects presented in different orientations. RTs increased as the angle of rotation increased from the ‘upright’ orientation (i.e. the orientation in which the object was most commonly experienced). Jolicoeur (1985) therefore suggested that the representation for most objects is viewer-centred and that the upright orientation is ‘canonical’. Similarly Palmer et al. (1981) suggested that an object can have a

Effect of Rotation 96 canonical perspective view which maximises the visibility of the object’s most salient features.

This suggestion that most objects have a canonical orientation is supported by neuropsychological studies. Turnbull et al. (in press) describe a patient who shows difficulties in establishing the canonical orientation of objects, though shows little impairment in recognising objects. This is independent of the patient’s ability to perceive orientation, since she is able to correctly carry out orientation matching tasks even matching the orientation of two objects.

Practice effects

Jolicoeur (1985) also found that after practice recognising all test orientations, subjects became less affected by the stimulus orientation and that the speed of recognition of all non-upright orientations was similar. Familiarity with an object in multiple orientations, therefore, diminishes both the effect of rotation on object recognition and the canonical nature of the upright orientation. This might be interpreted as the speed of mental rotation becoming faster with practice. Modification of the speed of mental rotation, however, did not generalise to untrained objects.

McMullen and Jolicoeur (1990; 1992) presented images of line-drawn objects and asked subjects to either name the object or to locate a dot which was placed near the top or the bottom of the object. A linear increase in RTs for both tasks was found when the image was rotated away from the upright orientation (for 0° to 120° rotation). The effect of orientation on object naming decreased as images were viewed more frequently. One can speculate that experience in recognising one object (at different orientations) could result in the establishment of a greater number of orientation specific (viewer-centred) templates for that object, and consequently the angular distance required for the matching process would be decreased, reducing all RTs. Nonetheless, such practice effects were less prominent when the subject had to indicate the top and the bottom of rotated objects (dot location task). In this task, the linear effect of stimulus orientation on top-bottom discrimination remained despite practice. McMullen and Jolicoeur (1992) therefore argued that an object-centred representation cannot be used to carry out the top-bottom discrimination task^. Instead, knowledge

^ Most object-centred representational schemes use the ‘major axis’ of an object as an internal spatial reference for describing the disposition of object parts (Marr and Nishihara, 1978; Biederman, 1987).

about the location of object features (or the orientation of the object's principle axis) relative to the viewer is critical for determining the top and bottom of objects.

To account for these effects one needs to differentiate recognition of whether an object is upright with respect to self or gravity and recognition of an object without regard to its orientation. These two tasks are differentiated by the studies of McMullen and Jolicoeur (1992).