Item

Abstract

Prior studies have shown that humans can recognize faces by touch alone but perform poorly in cross-modal face recognition [1]. Here we want to shed further light on haptic face recognition
with four experiments using a well-defined stimulus face space based on the morphable MPIFace-
Database. Experiment 1 used a same/different task with sequentially presented faces
which established that subjects were able to discriminate faces haptically, using short term
memory. In Experiment 2 we used an old/new recognition task for which different sets of
three faces (out of six) were learned haptically with three subsequent haptic test-blocks and
one visual test-block. In contrast to Casey and Newell (2007) we used the same printed face
masks for recognition in both modalities. We found that participants could recognize faces
haptically although recognition accuracy was low (65) and tended to decrease across blocks.
Cross-modal recognition, however, was at chance level (48). In Experiment 3, we changed
the design such that haptic memory was refreshed before each test-block by repeated exposure
to the three learned faces. We found that performance increased significantly to 76 and
that it became more consistent across blocks. Most importantly, however, we found clear
evidence for cross-modal transfer as visual performance rose above chance level (62). Our
results demonstrate that during visual face recognition, participants have access to information
learned during haptic exploration allowing them to perhaps form a visual image from haptic
information. In Experiment 4, we interchanged learning and recognition modality with respect
to Experiments 2+3, testing within-modality recognition in the visual domain and cross-modal
transfer by haptic recognition of the face masks. Using the same experimental design as in
Experiment 2, we found that performance in the visual within-modality condition increased
significantly to 89 and that it became more consistent across blocks (71 compared to 39
for Experiment 2). However, recognition accuracy decreased across blocks (from 96 to 87).
Interestingly, cross-modal performance was significantly higher than in Experiments 2 (at 69)
demonstrating a clear advantage in cross-modal transfer for vision as the learning modality. The
reasons for the observed differences in cross-modal transfer remain to be investigated. Possible
factors include differences in visual versus haptic memory permanence, vision as the dominant
and therefore preferred learning modality, and finally the role of visual imagery in cross-modal
transfer.