Abstract
Identification of subtle human facial expression is important in human interaction such as human-robot communication, robot safety and digital signage. In this paper, we propose a method to classify two types of facial expression; one is a neutral face (expressionless) and the other is subtle expression of happiness using an image acquired with a camera. In the proposed method, facial shape features such as wrinkle-angle or unevenness of face surface are extracted by using Gabor filters in multiple ROIs on a human face. Here, high angle-resolution Gabor filters can detect subtle facial expression. Positions of detection window and Gabor filter parameters are optimized by the learning process using AdaBoost. Facial expression changes are estimated by comparing an input image to learned database with the features. By the results of experiments using 324 real images, it has been confirmed that the performance of our system have 91% precision rate and 84% recall rate in the case of subtle changes. Here the processing time is within 210msec. and it is applicable to practical system.
