hide
Free keywords:
-
Abstract:
We investigated how body motion might influence the recognition of facial identity. Two heads from the MPI face database were attached to the same 3-D body model. These avatars were then animated with two different sequences of karate motions, one performed by a professional (A) and the other by an amateur (B). In a learning stage, observers (N=16) were trained to recognise these two animated figures. Test stimuli were created by morphing the learned heads in 10 steps from 0 (head A) to 100 (head B). Note that the 50 morph head is completely ambiguous. Each head was then attached to the same body and animated with motion sequences associated with either A or B. At test, observers were shown motion segments and had to decide whether the face looked more like A or B. Fitted psychometric functions were used to determine the 50 threshold. Observers consistently rated the 50 morphed head as avatar A when it was animated with A‘s motion and as avatar B when it was animated with B‘s motion. These results suggest that an observers‘ ability to determine the facial identity of an avatar can be biased by the way that avatar moves.