PROJECT TITLE :
A Novel Dynamic Model Capturing Spatial and Temporal Patterns for Facial Expression Analysis
Incorporating spatial and temporal patterns present in facial behavior should substantially improve facial expression analysis, however the patterns have not yet been fully employed. We address this problem by developing an interval temporal restricted Boltzmann machine (IT-RBM), a unique dynamic model capable of capturing both universal spatial patterns and complex temporal patterns in face behavior for facial expression interpretation.
A facial expression is thought to be a multifaceted activity made up of consecutive or overlapping primitive facial actions. A two-layer Bayesian network is used to represent these complex temporal patterns using Allen's interval algebra. The primitive face events are represented by the nodes in the upper layer, while the temporal links between those events are depicted by the nodes in the lower layer. A multi-value constrained Boltzmann machine, in which the visible nodes are facial events and the connections between hidden and visible nodes mimic intrinsic spatial patterns, also captures inherent universal spatial patterns in our model.
The authors present efficient learning and inference techniques. Experiments on posed and spontaneous expression differentiation and recognition show that our proposed IT-RBM outperforms current research by including these facial behavior patterns.
Did you like this research project?
To get this research project Guidelines, Training and Code... Click Here