Probabilistic Subpixel Temporal Registration for Facial Expression Analysis
- Cite this paper as:
- Sariyanidi E., Gunes H., Cavallaro A. (2015) Probabilistic Subpixel Temporal Registration for Facial Expression Analysis. In: Cremers D., Reid I., Saito H., Yang MH. (eds) Computer Vision -- ACCV 2014. ACCV 2014. Lecture Notes in Computer Science, vol 9006. Springer, Cham
Face images in a video sequence should be registered accurately before any analysis, otherwise registration errors may be interpreted as facial activity. Subpixel accuracy is crucial for the analysis of subtle actions. In this paper we present PSTR (Probabilistic Subpixel Temporal Registration), a framework that achieves high registration accuracy. Inspired by the human vision system, we develop a motion representation that measures registration errors among subsequent frames, a probabilistic model that learns the registration errors from the proposed motion representation, and an iterative registration scheme that identifies registration failures thus making PSTR aware of its errors. We evaluate PSTR’s temporal registration accuracy on facial action and expression datasets, and demonstrate its ability to generalise to naturalistic data even when trained with controlled data.
Supplementary material (avi 17,790 KB)