Happy Emotion Recognition From Unconstrained Videos Using 3D Hybrid Deep Features

Facial expressions have been proven to be the most effective way for brain recognize human emotions in a variety of contexts. With exponentially increasing research emotion detection recent years, facial expression recognition has become an attractive, hot topic identify various basic emotions. Happy is one such with many applications, which more likely recognized by than other measurement instruments (e.g., audio/speech, textual and physiological sensing). Nowadays, methods developed identifying multiple types emotions, aim achieve best overall precision all emotions; it hard them optimize accuracy single happiness). Only few are designed happy captured unconstrained videos; however, their limitations lie that processing severe head pose variations not considered, still satisfied. In this paper, we propose Emotion Recognition model using 3D hybrid deep distance features (HappyER-DDF) method improve utilizing extracting two different visual features. First, employ Inception-ResNet neural network long-short term memory (LSTM) extract dynamic spatial-temporal among sequential frames. Second, detect landmarks’ calculate between each landmark reference point on face nose peak) capture changes when person starts smile (or laugh). We implement experiments both feature-level decision-level fusion techniques three video datasets. The results demonstrate our HappyER-DDF arguably accurate several currently available models.