Facial Recognition under Expression Variations

Researchers in different fields such as image processing, neural sciences, computer programs and psychophysics have investigated number of problems related to facial recognition by machines and humans since 1975. Automatic recognition of the human emotions using facial expression is an important, but difficult problem. This study introduces a novel and automatic approach to analyze and recognize human facial expressions and emotions using a Metaheuristic Algorithm (MA), which hybridizes iterated local search and Genetic Algorithms with Back-Propagation algorithm (ILSGA-BP). Back Propagation algorithm (BP) was used to train and test the extracted features from the extracted right eye, left eye and mouth using radial curves and Cubic Bezier curves, M4 was used to enhance and optimize the initial weights of the traditional BP. FEEDTUM facial expression database was used in this study for training and testing processes with seven different emotions namely; surprise, happiness, disgust, neutral, fear, sadness and anger. A comparison of the results obtained using the extracted features from the radial curves, Cubic Bezier curves and the combination of them were conducted. The comparison shows the superiority of the combination of the radial curves and the Cubic Bezier curves with percentage ranges between 87% and 97% over the radial curves alone with a percentage ranges between 80% and 97% and over the Cubic Bezier curves with a percentage ranges between 83% and 97%. Moreover, based on the extracted features using the radial curves, Cubic Bezier curves and the combination of them, the experimental results show that the proposed ILSGA-BP algorithm outperformed the BP algorithm with overall accuracy 88%, 89% and 93.4% respectively, compared to 83%, 82% and 85% respectively using BP algorithm.