The Fourth Asian Conference on Vision (ACV2006)i]j

Recognition of Facial Identity and Facial Expression from Spatial Band-Pass Filtered Stimuli

Mikio Inagaki1, Maiko Hata2, Ichiro Fujita1

1Graduate School of Frontier Biosciences, Osaka University, Osaka, Japan
2School of Engineering Science, Osaka University, Osaka, Japan

1. Introduction
People can recognize several different aspects of face, such as identity, expression and gaze direction, from a face image. Neuropsychological evidence indicates that these functions can be dissociated from each other and, thus, may be mediated by different neural mechanisms. In order to gain clues of these mechanisms, we examined how different spatial frequency bands of face images contribute to discrimination of facial identity and facial expression.

2. Methods
Subjects (n=10) were required to perform a delayed matching-to-sample task. In each trial, a sample face stimulus was first shown for 1 s, and then four target face stimuli including the sample face appeared for 5 s after an interval of 100 ms. The task of the subjects was to select the target face identical to the sample face. The sample stimulus was selected from a set of 48 face images of 4 persons with 4 facial expressions (neutral, happy, fear, anger), unfiltered or spatial band-pass filtered (4 cycles/image or 16 cycles/image). The target stimuli were always unfiltered. In identity discrimination tasks, the target stimuli were faces of 4 different persons with the same facial expression. In expression discrimination tasks, the target stimuli were faces of 1 person with 4 different facial expressions. We systematically manipulated the task difficulty by changing the contrast variance of the sample images.

3. Results
We plotted correct performance rates against contrast variance, and obtained psychometric functions separately for identity and expression discrimination tasks with different spatial filtering conditions. Under unfiltered and 16-cycles/image filtered conditions, the contrast threshold for the facial expression task was lower than that for the facial identity task. No such tendency existed for 4-cycles/image filtered conditions.

4. Conclusion
The results suggest that people are more sensitive to facial expression than to facial identity under low contrast conditions. Relatively high spatial frequency bands of face images contribute to this sensitivity to facial expression.

This work was supported by a MEXT grant (17022025).
Correspondence: Ichiro Fujita, Ph. D. (