| Popis: |
Automatic recognition of human facial gestures has recently attained a significant place in multimodal human-machine interfaces and further applications. This paper presents our proof-of concept for Facial Expression Interpretation, which is real-time capable and robust enough to be deployed to real-world scenarios. We exploit model-based techniques and adapt them specifically to facial expression recognition. First, we extract informative features about faces from the image data that describes skin color and lip color regions. Furthermore, we learn the objective function for model fitting from example images as opposed to constructing it manually. The obtained system shows promising referring to its recognition results and we have successfully conducted live demonstrations that were attended by industrial and political audience, such as in [11]. In future work, we aim at integrating multi-modal feature sets, for which we have already preliminary results [21]. Furthermore, we will present structural, motion and textural features to the classifier. We will apply Early Sensor Fusion in order to keep all knowledge for the final decision process and to exploit the ability of a combined feature-space optimization. |
