Conclusion

One of major stumbling blocks in expression modeling from unconstrained data has been inaccurate alignment and normalization of face images. Thus, models were built using constrained and unnatural data which is not scalable to real life cases. In this paper we proposed an expression modeling technique that can work robustly with natural data and have high recognition accuracy. Our models are built on feature based on motion motion field histograms, which have robustness against errors in rotation, translation and scale changes during image alignment. The results demonstrate a 44% average performance increase over traditional optic flow methods for expressions extracted from unconstrained interactions.