Differential Evolution to Optimize Hidden Markov Models Training: Application to Facial Expression Recognition

The base system in this paper uses Hidden Markov Models (HMMs) to model dynamic relationships among facial features in facial behavior interpretation and understanding field. The input of HMMs is a new set of derived features from geometrical distances obtained from detected and automatically tracked facial points. Numerical data representation which is in the form of multi-time series is transformed to a symbolic representation in order to reduce dimensionality, extract the most pertinent information and give a meaningful representation to humans. The main problem of the use of HMMs is that the training is generally trapped in local minima, so we used the Differential Evolution (DE) algorithm to offer more diversity and so limit as much as possible the occurrence of stagnation. For this reason, this paper proposes to enhance HMM learning abilities by the use of DE as an optimization tool, instead of the classical Baum and Welch algorithm. Obtained results are compared against the traditional learning approach and significant improvements have been obtained.