非表示:
キーワード:
-
要旨:
Human visual encoding of scenes was studied with respect to view specificity:
the extent to which observers ability to recognize a familiar scene depends on
the limited set of directions experienced during training. Precise control of
cues was achieved by using a computer model of a virtual room. A novel
explorative search paradigm was implemented using real-time image generation
and provided a controlled yet natural means of familiarisation.
After training, observers where able to recognize both familiar and novel
direction views but the latter involved more errors and required more
processing time. Observers were also able to identify the corresponding
floorplan indicating that the encoding could provide abstract relation
information.
In subsequent tests in which we replaced the interactive training stage with
passively observed sequences of snap-shots we found that although
familiar-view recognition persisted the novel view generalisation performance
diminished. This was so, even when binocular disparities were used to provide
more depth information.
These results suggest that mental encoding of scenes is view-based although
the observed generalisation to novel views cannot readily be explained by
simple transformations on 2D stored views. Furthermore, our study highlights
the importance of allowing natural behaviour during visual familiarisation
tasks.
