Facial expression analysis using nonlinear decomposable generative models.Image.Marked资料.pdf

Facial Expression Analysis using Nonlinear Decomposable Generative Models Chan-Su Lee and Ahmed Elgammal Computer Science, Rutgers University Piscataway NJ 08854, USA {chansu, elgammal}@cs.rutgers.edu Abstract. We present a new framework to represent and analyze dynamic facial motions using a decomposable generative model. In this paper, we consider facial expressions which lie on a one dimensional closed manifold, i.e., start from some configuration and coming back to the same configuration, while there are other sources of variability such as different classes of expression, and different peo- ple, etc., all of which are needed to be parameterized. The learned model supports tasks such as facial expression recognition, person identification, and synthesis. We aim to learn a generative model that can generate different dynamic facial appearances for different people and for different expressions. Given a single image or a sequence of images, we can use the model to solve for the tempo- ral embedding, expression type and person identification parameters. As a result we can directly infer intensity of facial expression, expression type, and person identity from the visual input. The model can successfully be used to recognize expressions performed by different people never seen during training. We show experiment results for applying the framework for simultaneous face and facial expression recognition. Sub-categories: 1.1 Novel algorithms, 1.6 Others: modeling facial expression 1 Introduction The appearance of a face performing a facial expression is an example of a dynamic appearance that has global and local deformati