ausblenden:
Schlagwörter:
-
Zusammenfassung:
How is basic level categorization achieved in the human brain? Deforming shape (morphing) transformations are
well suited to describe the shape variability of members of common basic level categories (Figure 1). Behavioral
experiments showed that categorization performance deteriorates systematically with increasing amount of morph
transformation (Dahl, Graf, and Bülthoff, 2003; Graf, 2002). A possible explanation for these findings is that
categorization relies on time-consuming compensation processes (deformable template matching). We tested,
using fMRI, whether spatial (deforming) compensation processes are involved in categorization, and
hypothesized that such processes would additionally activate the dorsal stream.
Objects from 25 common basic level categories were generated by morphing between two members of the same
category (using 3ds max). Subjects were familiarized with all objects before the experiment. Eleven subjects
participated in two tasks, starting with the categorization task. Subjects had to decide as fast as possible whether
two sequentially presented objects belonged to the same basic level category. The transformational distance
between category members was varied (event-related design). In a second task, the same observers perceived
intact morphing sequences, scrambled morphing sequences, and static presentations of different morph exemplars
(updating task, block design). fMRI data were acquired on a 3T scanner (Siemens Trio), with 24 slices of 64x64
voxels every two seconds (resolution of 3x3x5 mm3), and whole brain coverage.
Behaviorally, we found that in the categorization task, the response latencies for same trials increased with
increasing morph distance between two category members. Correspondingly, the fMRI analysis, contrasting long
vs. short morph distance, revealed an increasing BOLD signal in LOC (lateral occipital complex) in both
hemispheres. Moreover, we found activation in the superior parietal cortices (BA 7) and in the right frontal cortex
(BA 44) (Figure 2). Control analyses showed that this pattern of activation cannot be reduced to task difficulty, or
increasing dissimilarity between the objects. In the second task we found dorsal activation for the comparison
between intact vs. scrambled morphing sequences (and additional bilateral activation in LOC when intact
morphing sequences were compared to scrambled and static sequences). The parietal activation, which was
located close to the dorsal activation in the categorization task (Figure 3), seems related to the perception of
continuously morphing objects.
These results suggest that basic level categorization is not limited to the ventral pathway, but rather relies on a
network of ventral, dorsal and frontal activation. The activation within this network systematically depends on the
amount of shape transformation. The dorsal activation seems related to compensational processes taking place in
parietal cortex, i.e. spatial (deforming) transformation processes. These findings are in accordance with an
alignment approach of object recognition and categorization.
