Recognition of Change in Facial Expressions From Video Sequence Using Incenter-Circumcenter Pair Gradient Signature

Facial expression recognition system provides us enormous scope for studying characteristics of different human facial expression. Basic expressions (anger, disgust, fear, happiness, sadness, and surprise) can be distinguished from one another with minimum overlapping features between them. A human facial emotion can be represented uniquely if we are able to capture such features having high discrimination power. In this paper, we propose a method for recognizing facial expression by using salient landmark-induced gradient signature derived from the line segment joining circumcenter and incenter pair of a triangle obtained by joining relevant landmark points. We use appearance-based models to detect essential landmark points on face. We are considering only those landmark points situated at eyes, lips, eyebrows, and nose to get geometric feature. This landmark-based geometric facial feature has been used to describe changes in emotion from neutral through basic expression. We examine our proposed method on Chon-Kanade (CK+), MMI, and MUG benchmark image sequence databases and compare its performance under different motion image databases. In our experiment, dynamic behavior of every individual expression has been recognized with impressive recognition rate of accuracy.