Pose-variant facial expression recognition using an embedded image system

Kai-Tai Song*, Meng Ju Han, Shuo Hung Chang

*Corresponding author for this work

Research output: Contribution to journalConference articlepeer-review


In recent years, one of the most attractive research areas in human-robot interaction is automated facial expression recognition. Through recognizing the facial expression, a pet robot can interact with human in a more natural manner. In this study, we focus on the facial pose-variant problem. A novel method is proposed in this paper to recognize pose-variant facial expressions. After locating the face position in an image frame, the active appearance model (AAM) is applied to track facial features. Fourteen feature points are extracted to represent the variation of facial expressions. The distance between feature points are defined as the feature values. These feature values are sent to a support vector machine (SVM) for facial expression determination. The pose-variant facial expression is classified into happiness, neutral, sadness, surprise or anger. Furthermore, in order to evaluate the performance for practical applications, this study also built a low resolution database (160x120 pixels) using a CMOS image sensor. Experimental results show that the recognition rate is 84% with the self-built database.

Original languageEnglish
Article number71300X
JournalProceedings of SPIE - The International Society for Optical Engineering
StatePublished - 1 Dec 2008
Event4th International Symposium on Precision Mechanical Measurements - Hefei/Jiuhua Mountain, Anhui, China
Duration: 25 Aug 200829 Aug 2009


  • Active appearance model
  • DSP-based vision system
  • Emotion recognition
  • Human-robot interaction
  • Pose-variant facial expressions

Fingerprint Dive into the research topics of 'Pose-variant facial expression recognition using an embedded image system'. Together they form a unique fingerprint.

Cite this