Improved Group Facial Expression Recognition Using Super-Resolved Local Facial Multi Scale Features

Group based emotion recognition is attracting more and more researchers due to its wide range of applications from social to political aspects. Emotion recognition plays a vital role in handling gatherings of people, group activity, and crowd in a better and smoother way to avoid any conflicting incident. Group level approaches are split into two basic categories: top down and bottom-up approaches. The bottom-up approach considers local level features such as facial expression, color, edges and trajectory of individuals, while the top-down approach infers global information like group and sub group information, age, gender, the effect of the scene. The proposed approach in this paper solely relies on extraction of the local level facial expression to perform emotion recognition. In order to extract contextual information from the images, we utilized super resolution and multi scale based feature extraction techniques. The proposed approach is found satisfactory and achieved very good and comparable accuracy in spite of utilizing only local facial features. The proposed approach achieves overall 67% accuracy on standard Group affect database (GAF) 3.0.