A new model and process architecture for facial expression recognition

In this paper we address the problem of facial expression recognition. We have developed a new facial model based only on visual information. This model describes a set of bidimensional regions corresponding to those elements which most clearly define a facial expression. The problem of facial gestures classification has been divided into three subtasks: face segmentation, finding and describing relevant facial components and, finally, classifying them into one of the predefined categories. Each of these tasks can be solved independently using different techniques already applied to a wide range of problems. This have led us to the definition of a modular, generic and extensible process architecture. A prototype has been developed which makes use of different simple solutions for each module, using a controlled environment and a low-cost vision system. We report the experimental results achieved by the prototype on a set of test images.