Abstract
It is known that humans can recognize familiar faces at relatively low resolution (e.g., 16x16 pixels or 8 to 16 cycles per face), indicating a transfer of face learning from high-to low-resolution images. It is not known, however, whether learning blurry faces is followed by better recognition of high-resolution faces or blurry faces. This is particularly relevant to people with low vision who wish to recognize faces learned prior to the onset of eye disease and who also wish to learn new faces in the presence of reduced acuity. It may also be relevant in normal vision when learning faces of people at a distance (near the acuity limit) and then recognizing them close up. We had normally-sighted subjects learn a set of 20 new faces with and without blur. During the learning phase, each face was presented for 2 seconds. After a 2-minute retention interval, subjects' recognition performance was measured with and without blur using a temporal 2AFC-paradigm. Learning and test conditions were counterbalanced across subjects. A set of unfamiliar Korean characters was also learned using the same procedure. We found that blurred faces are recognized better than high-resolution faces if face learning occurred in the presence of blur (81% vs. 69% accuracy; t(9) = 2.57, p = 0.03). However, the same pattern did not occur for unfamiliar Korean printed characters, that is, even when the characters were learned with blur, recognition was no better for blurry characters than for high-resolution characters (77% vs. 74%; t(9) = 0.44, p = 0.67). Perhaps, the blur used for this study was not severe enough to affect the visual characteristics used to learn and recognize Korean characters. Our findings suggest that the visual characteristics which are useful for learning and recognizing blurred faces may not be the same as those for learning and recognizing high-resolution faces.
This work was supported by NIH grant R01 EY002934.