An efficient face model for facial expression recognition

Existing facial expression recognition (FER) algorithms aim to extract discriminative features from a face. These discriminative features can be extracted only from the informative regions of a face. In this view, several face models are proposed which are mainly intended to extract geometrical features from a face, and hence these models may not be suitable for extract discriminative texture features from a face. We proposed a novel face model based on projection analysis of a face. Our proposed projection analysis evaluates the distribution of informative regions of a face. This is done by projecting the expressive face images onto their corresponding neutral images. Hence, the proposed face model can efficiently extract distinctive texture features from a face. Additionally, the proposed face model can extract geometrical features as well. The performance of the proposed face model is evaluated on MUG datasets which shows that the proposed face model outperforms several existing face models. Also, the proposed face model can give a recognition accuracy of 97.3% which is significantly better than the performance of state-of-the-art face models.