Enhancing Emotion Recognition in Images Through Context-Aware Deep Learning

Abstract

Computer vision, psychology, and neuroscience are among the disciplines where the study of emotional recognition from images is developing. Notwithstanding developments in computer vision, emotion detection faces several challenges: insufficient accuracy resulting from facial-centric models trained on limited datasets and the absence of properly contextualised picture collections necessary for capturing the real complexity of human emotions. This study suggests a new method for contextual emotional detection by aggregating physical experiences with environmental signals. We built LEMOTIC, a large and varied data set that guarantees multiple emotional settings. We built our approach using deep convolution neural network (DCNN) architecture. To better grasp emotions in context, we bridge the gap between 26 separate emotional categories and the three continuous emotional dimensions: valence, arousal, and dominance. Our model demonstrated its capacity to identify emotions from full-body postures and environmental signals by astonishingly high performance after rigors hyperparameter tuning. Our Sentiment_Recognition_Model outperforms previous methods on continuous and discrete emotional variables in accuracy and generalisation. Our research shows that contextual and somatic emotions are crucial in studying emotions in images. Our method uses deep learning and a well-created dataset to improve context-aware emotion identification, enabling more precise applications in social robots, affective computing, and human-computer interaction.