Facial Expression Recognition Using Uniform Local Binary Pattern with Improved Firefly Feature Selection

Facial expressions are essential communication tools in our daily life. In this paper, the uniform local binary pattern is employed to extract features from the face. However, this feature representation is very high in dimensionality. The high dimensionality would not only affect the recognition accuracy but also can impose computational constraints. Hence, to reduce the dimensionality of the feature vector, the firefly algorithm is used to select the optimal subset that leads to better classification accuracy. However, the standard firefly algorithm suffers from the risk of being trapped in local optima after a certain number of generations. Hence, this limitation has been addressed by proposing an improved version of the firefly where the great deluge algorithm (GDA) has been integrated. The great deluge is a local search algorithm that helps to enhance the exploitation ability of the firefly algorithm, thus preventing being trapped in local optima. The improved firefly algorithm has been employed in a facial expression system. Experimental results using the Japanese female facial expression database show that the proposed approach yielded good classification accuracy compared to state-of-the-art methods. The best classification accuracy obtained by the proposed method is 96.7% with 1230 selected features, whereas, Gabor-SRC method achieved 97.6% with 2560 features.